Alternative Strategies to Reduce Maternal Mortality in India: A Cost-Effectiveness Analysis

A cost-effectiveness study by Sue Goldie and colleagues finds that better family planning, provision of safe abortion, and improved intrapartum and emergency obstetrical care could reduce maternal mortality in India by 75% in 5 years.

versus urban areas) can be accurately represented in terms capacity and cost. Facility levels (India Facility Survey Phase-II [2003]) are categorized as (1) primary-level facilities, which may not have all bEmOC functions but could function as birthing centres with SBA staffing, 24-hour intrapartum care, and reliable referral connections (e.g., subcentre, primary health centre [PHC]); (2) secondary facilities with bEmOC capacity (e.g., first referral unit [FRU], community health centre [CHC]; and (3) tertiary facilities with cEmOC capacity (e.g., district hospital, some first referral units) [India Facility Survey Phase II, 2003]. We recognize that some tertiary sites will not have a blood bank and some secondary sites may eventually be able to perform c-section; further, we recognize that in the strategies that include stepwise investments in infrastructure and facility improvements, not all facilities will be expected to be fully implemented as one of the three distinct types. However, because the costs, functions and staffing are fairly closely aligned with basic or comprehensive EmOC capacity, this simple categorization captured the most important dimensions for purpose of this analysis. Above is a stylized example of how public health facilities in India, as categorized in the India Facility Survey Phase-II [2003], may be superimposed on our general model framework.

5
All models were built using TreeAge Pro 2008 (TreeAge Software Inc., Williamstown MA) and analyzed using IBM/Lenovo Dual-Core VT Pro Desktop computers running Microsoft Windows XP, using Microsoft Excel 2007 and Visual Basic for Appplications 6.5 (Microsoft Corp., Redmond WA). We used Monte Carlo simulation to generate the number of per woman events such as pregnancies, live births, facility-based births, and maternal complications. This output is useful for both calibration exercises, as well as for assessing internal consistency and projective validity of the model by generating outcomes in similar formats to clinical studies. We use second order Monte Carlo simulation to assess parameter uncertainty.

Subsection A Data and Assumptions
Age-specific probability of pregnancy To estimate a fertility rate in the absence of any family planning, we use data from Afghanistan [Amowitz 2002, Bartlett 2005, where contraceptive use is low, the maternal mortality ratio (MMR) is among the highest in the world (~1,600 deaths per 100,000 live births), and access to modern health care is limited (< 5%) [AbouZahr 2004]. We synthesized data on family planning, abortion, and demographics (i.e., crude birth rate in Afghanistan of 308 pregnancies per 1,000 population) to approximate an average annual natural fertility rate of 31% [PRB 2004].The model allows for agespecific inputs for fertility and use of contraception. We assumed 15% of all pregnancies end in spontaneous abortion, of which approximately one-third result in incomplete abortion requiring medical intervention [Harlap 1980, Menken 2006]. We assumed women with long-term complications such as infertility or untreated obstetric fistula did not become pregnant again. We assumed women with complications that were treated (e.g., severe anemia, surgically repaired fistula) could become pregnant again.

Anemia
In the National Family Health Survey-3 (2005Survey-3 ( -2006, among women of reproductive age, the overall prevalence of anemia was 52%, with mild anemia affecting 35%, moderate anemia affecting 15%, and severe anemia affecting 2%. [IIPS 2007] In Uttar Pradesh (UP), based on data from the District Level Household & Facility Survey of the Reproductive & Child Health Project (DLHS-RCH phase-2, round-2 survey), anemia accounted for 55% of all indirect causes of death, and 15% of all maternal deaths. [Mills 2007] The relative risk of death from maternal complications is 3.5 times greater with severe anemia and 1.35 times greater with moderate anemia, compared to a woman without anemia. [Brabin 2001] Antenatal care presents an opportunity to detect and treat anemia. We assumed severe and moderate anemia were associated with a higher relative risk of death from pregnancy-and delivery-related complications, although anemia differentially affected mortality from postpartum hemorrhage and sepsis, and complications following unsafe abortion. We conservatively assumed that severe anemia did not impact the case fatality rate of untreated obstructed labor.

Adjustment of case fatality rates for heterogeneity in severity and co-morbidity
Baseline estimates for cause-specific case fatality rates were from a review conducted by the Disease Control Priorities Project (DCP2). ] -maternal hemorrhage (1%), sepsis (1.3%), hypertensive disorders of pregnancy (1.7%), obstructed labor (0.7%) ]. These case fatality rates are lower than those reported in West Bengal, India which ranged from 0.9% to 3.5%, implying they could be underestimates [Biswas 2005]. We assumed that some of the variation reported in the literature is attributable to the heterogeneity in severity, and estimates based on small sample sizes. Case fatality rates were thus adjusted based on complication severity (e.g., life threatening complications requiring cEmOC) and underlying severity of anemia.
We assumed that life-threatening complications requiring cEmOC-level services were associated with a higher case fatality rate (in the absence of treatment) than that of non-life-threatening complications. This relative risk was determined through a number of calibration exercises in which the model was first parameterized using the best natural history data available, then adjusted to reflect the current standard of care in India, and finally by allowing this relative risk to vary such that the model fit multiple epidemiologic targets simultaneously, including the MMR, life expectancy, total fertility rate, and distribution of maternal mortality causes. We additionally assumed an increased relative risk of mortality secondary to severe maternal hemorrhage, sepsis and abortion, ranging from 1.5 to 3.0 with severe anemia [Brabin 2001] Assumptions for the proportion requiring bEmOC and cEmOC are described below.
Among the estimated 15% of pregnant women in developing countries who experience pregnancyrelated complications, 7% require care at centers with surgical capacity (cEmOC) and 2-3% will require surgery [Johns 2007]. Initial estimates of the proportion of complications requiring basic versus comprehensive EmOC care were derived from a study providing WHO expert opinionbased estimates, for each maternal complication, of the proportion of complications that will require surgical intervention, blood transfusion, or management of shock [Johns 2007].
Johns ( In setting-specific models (e.g., rural versus urban India), these proportions were slightly altered either using data about the availability of certain levels of care in basic and comprehensive facilities and extent of training of personnel, or from insights gained during the calibration exercises. (Subsection B and C). In certain cases, we had specific data to assist with modifying estimates. For example, Johns et al assumed 85.8% of eclampsia cases (HTN) require bEmOC, while 14.2% require cEmOC [Johns 2007]. This estimate is similar to an earlier study which showed that severe pre-eclampsia and eclampsia required treatment with intravenous hydralazine and magnesium sulfate and in addition, approximately 10% of all cases were assumed to require emergency cesarean section [Cahuana-Hurtado 2004]. Initial assumptions (and ranges) used in India with respect to severity and need for basic versus comprehensive EmOC are provided below; sensitivity analyses were conducted to assess the implications of using the upper and lower bounds.

Hemorrhage Incidence and case fatality rate
In a systematic review of 34 datasets, representing over 35,000 maternal deaths, Khan et al. [2006] found hemorrhage to be the leading cause of death throughout the world, accounting for a range of 1.4% to 49.6% of all maternal deaths, and the cause of the highest proportion of deaths in Asia and Africa. In India, hemorrhage is the leading cause of maternal mortality and contributes up to 23% of all maternal deaths (range, 7-30%). [Khan 2006] Maternal hemorrhage is categorized according to its timing in relation to delivery: antepartum, intrapartum, or postpartum. The etiologies and management of maternal hemorrhage differ among these three categories.
Initial estimates for the overall incidence of PPH were based on data from the WHO's Global Burden of Disease study (2002) specifically for the SEAR region. We calculated estimates for the incidence of PPH (0.097625) using data on the number of cases (n=3,692,000) [WHO 2007a] and the total number of births (n =37,820,000) [UNICEF 2004]. The risk of PPH was modified to reflect assumptions in the WHO's "Global burden of maternal hemorrhage in the year 2000" [Dolea 2003a], specifically that the incidence of PPH, defined as > 1000ml of blood loss in the oxytocin arm, was 2.85% within 1 hour postpartum in women who were actively managed, as estimated from the MISO trial. [Dolea 2003a] We assumed that the incidence of PPH in women who are managed expectantly by a skilled birth attendant would be twice as high as found in the MISO trial, or 5.7% of births. [Chandhiok 2006, Dolea 2003a, and that births without skilled attendance would be twice as high as those with skilled birth attendance. [Alfirevic 2007, Derman 2006, Gulmezoglu 2007, Hoj 2005] Extrapolating from these data, we assumed all births in a facility with emergency obstetrical care would be actively managed with a 2.85% risk of PPH, all other births with a skilled attendant would be expectantly managed with a 5.7% risk of PPH, and all births attended by a family member or traditional birth attendant, or when delivery was alone, would be associated with an 11.4% risk of PPH. This approximates the range of 8% to 15% reported in the literature.
The initial estimate for the case fatality rate (CFR) was from a review conducted by the Disease Control Priorities Project (DCP2) ] who reported an average CFR of 1%; this estimate was adjusted according to case severity and underlying morbidity (e.g., severe anemia) by calibrating the model to fit multiple epidemiologic targets simultaneously. See section above on Adjustment of case fatality rates for heterogeneity in severity and co-morbidity. The implications of our adjusted CFR widened the implied plausible range. While the literature based range based on 0.01 (1%) average CFR from DCP2 was 0.007 -0.013, the expanded plausible range based on our adjustment of a CFR of 0.023 (2.2%) was 0.007 -0.030. Model projected mortality due to maternal hemorrhage, as well as MMR, TFR, and calendar deaths for 2005, closely approximated the empiric data.
To account for the uncertainty in our initial estimates, we established a plausible range for all the above parameters based on our literature review. There have been multiple studies, including 8 systematic reviews, of the incidence and case fatality rate of maternal hemorrhage. A Cochrane review demonstrated that active management with oxytocin results in a relative risk of 0.33 for blood loss >1000ml within the first 24 hours, compared to expectant management. [Prendiville 2000] This review has since been withdrawn due to concerns regarding the validity of these findings. [Prendiville 2009] An updated analysis is currently underway to ensure the use of more recent data. A Cochrane review comparing oxytocin to no uterotonics found an overall reduction in blood loss > 1000ml of 39% (RR 0.61). [Cotter 2001] Comparing oxytocin to no uterotonics when active management was used in both trial arms, the same review found a relative risk of 0.33, and when expectant management was common to both trial arms, a relative risk of 0.73 was found. [Cotter 2001] A study published in 2002 comparing active management with expectant management found a relative risk of 0.8 for blood loss > 500ml. [Geelhoed 2002] While severe PPH was not an endpoint captured in this trial, we would expect a slightly lower RR for blood loss > 1000ml. A more recent randomized control trial of a small number of women who gave birth at a maternity unit in Iran comparing active management with oxytocin to expectant management found conflicting evidence regarding the optimal method by which to manage the third stage of labor. [Kashanian In press] This trial found that active management did not decrease blood loss during the third stage of labor but did decrease the duration of this stage. Active management was associated with increased blood loss during the fourth stage of labor. Severe PPH was not an endpoint captured within this analysis.

11
In a systemic review of randomized trials, the prophylactic administration of oxytocin reduced severe PPH from 7% to 4.3% (RR 0.61), and that the relative risk of severe PPH when using misoprostol compared to placebo (when an outlier study was excluded) was 0.77 for 600 mg and 0.54 for 400 mg. Hofmeyr (2008) There was wide variation in the effect of misoprostol but all studies showed some effectiveness compared to placebo. A systemic review covering the period 1997-2006 (Carroli 2008) included 24 studies from the WHO database covering a period of 1997-2002 and an additional 166 reports assessed (from 2003-2006) with 14 included (total = 38) for a total of 224 datasets. These were stratified to those that reported PPH (n=120) and those that reported severe PPH (n=70), with severe PPH being defined as blood loss >1,000 ml. Overall quality was deemed adequate for ~47% of PPH datasets and ~59% of SPPH datasets. Overall prevalence of PPH was 6.09% (CI 6.06 to 6.11) with 10.55% when the blood loss was measured objectively. Overall prevalence of severe PPH was 1.86% (CI 1.82 to 1.90) with 3.04% when the blood loss was measured objectively. A high degree of heterogeneity was reported even in subgroups with similar characteristics. Severe PPH was reported at 3.84% for expectant management alone

Reduction in mortality
The incidence of maternal hemorrhage is dependent on the delivery setting, the use of expectant or active management, and the use of misoprostol. The model represents a range of approaches to reduce mortality from PPH: first, on the basis of delivery setting and use of expectant or active management of labor; second, by successful referral and access to quality care in an appropriate facility with basic or comprehensive emergency obstetrical care; third, with temporizing measures such as the antishock garment to reduce blood loss and shock en route to an adequate facility; fourth, by the use of misoprostol after PPH at home or in a birthing centre, to reduce total blood loss; and fifth, by the use of misoprostol in the community setting, at home or in a birthing centre or subcentre, to prevent PPH. Data for the first, second and third options are far more readily available than the fourth and fifth options, which reflects our choice to focus on these for the present analysis. We include one exploratory analysis of community-based SBA-administered misoprostol. We assumed optimal treatment of maternal hemorrhage in an appropriate facility with EmOC capacity consisted of intramuscular or intravenous oxytocin immediately after delivery, uterine massage, repair of any perineal or vaginal tears, and fluid replacement or blood transfusion [Cahuana-Hurtado 2004]; consistent with assumptions made by Adam (2005) and  we assumed an average reduction of 75% in the CFR [Adam 2005]. We varied this estimate from 60% to 90% for optimal management (i.e., bEmOC or cEmOC as necessary for severity).

Misoprostol
For the exploratory analysis evaluating the use of SBA-administered misoprostol at home and in birthing centres we varied effectiveness to prevent PPH from 10% to 75%, and conducted a baseline analysis using 25% and 50%. Among most of the studies that have assessed misoprostol, there is variation in methods and clinical practice (e.g., measurement of blood loss, management of the third stage of labor, and use of uterotonics to prevent versus manage PPH), which makes direct comparison difficult. In , oxytocin was reported to be more effective than misoprostol in reducing acute and severe PPH. However, the study included several developed countries and when distinguishes the data from just the developing countries, the risks are similar for oxytocin and misoprostol. More recent studies showed more positive results in terms of measured blood loss, versus placebo, for prevention (Derman 2006, Hoj 2005 of PPH. Derman et al (2006) reported the results of a placebo-controlled trial in rural India with auxiliary nurse midwives at home or in village subcentres who administered misoprostol 600 mcg orally or a placebo, in the context of expectant management. They found a statistically significant reduction in risk for acute PPH, severe PPH, additional uterotonics, transfer to higher level facility, and transfusion in the misoprostol group versus the placebo group. Hoj et al. (Hoj 2005) in Guinea-Bisseau, another large study (n > 600), showed a reduction in the risk for severe PPH although not acute PPH. The incidence of acute PPH (blood loss > 500) was high for both misoprostol and placebo groups (45% and 51%) as was the incidence of severe PPH (11% and 17%).  reported results of a study in the Gambia comparing TBA-administered 600 mcg misoprostol to 2 mg ergometrine (n> 1000), in which both drugs had similar risks of acute PPH (11-12%) and severe PPH (0.3-0.7%), and lower blood loss compared to placebo. Chandhiok et al. (Chandhiok 2006) in rural India with n > 1000 which compared 600 mcg misoprostol orally to methergine (both IM and PO) and showed similar risks of acute PPH for both groups which were very low (0.70-0.8%).
Our review also found that providing misoprostol as a prophylaxis (as a dose of 600ug) was more effective than placebos at preventing PPH in community births, having a relative risk of 0.59, but was not as effective in a hospital setting where it had a relative risk of 1.23. [Alfirevic 2007] This was reinforced by a Cochrane Review [Gulmezoglu 2007] which concluded that while misoprostol was less effective than oxytocin and associated with higher rates of shivering and fever, it showed promising results when compared to the placebo. A review of the evidence supporting guidelines found that for prevention of PPH, active management reduced risks during the third stage of labor.
It also found that misoprostol should only be used if oxytocin is unavailable [Leduc 2009]. This was supported by a study [Gulmezoglu 2009] which, cited a previous study by Prendiville et al. [2000], found that active management could reduce the risk of severe PPH by between 60% to 70%.
14 Hofmeyr (2008) found in a systematic review of randomized trials that there was no difference in severe morbidity between misoprostol and other conventional uterotonics, although those who were given misoprostol did experience more side effects. When an individual outlier study was excluded, results showed less blood loss with misoprostol than with the placebo provided during the trial; specifically, the prophylactic administration of oxytocin reduced severe PPH from 7% to 4.3% (RR 0.61), and that the relative risk of severe PPH when using misoprostol compared to placebo was 0.77 for 600 mg and 0.54 for 400 mg. This review did not answer the question of whether the relative mortality reduction owing to misoprostol preventing PPH-related deaths was offset by an increase in mortality caused by the drug. In addition, the review did not compare active versus expectant management [Hofmeyr 2009]  Johns et al (2007) estimates antepartum hemorrhage requiring management will complicate 2.2% of pregnancies and more specifically 0.11% pregnancies will be complicated by antepartum hemorrhage requiring caesarean section and 0.726% pregnancies will be complicated by antepartum hemorrhage requiring transfusion. As with postpartum hemorrhage, the course of antepartum hemorrhage can be unpredictable and a recurrent bleed can occur at any time and any severity level. Antepartum hemorrhage was not considered as a separate category in our model because we felt there were insufficient data on its epidemiology, natural history, and the impact of interventions in developing countries. [For these same reasons, estimates of death and disability attributable to antepartum hemorrhage were not included in the WHO's global burden of maternal hemorrhage]. That being said, we did calibrate to observed data on the distribution of deaths by cause in India, including PPH of all causes. [see Subsection C.]

Inclusion of antepartum hemorrhage
In contrast to postpartum hemorrhage, for which WHO region-specific incidence and mortality rates are available, the frequency of antepartum hemorrhage has been difficult to establish at the population level in developing countries due to a lack of: (1) widely accepted diagnostic criteria for this condition and (2) reliable ascertainment, which is grossly affected by the quality and availability of maternal care. Empirical data regarding the natural history of antepartum hemorrhage are also lacking. For example, the proportion of antepartum hemorrhages that present as severe or lifethreatening is unknown, as is the proportion of cases that ultimately require transfusion and/or cesarean section. The percentage of cases of antepartum hemorrhage that resolve only to recur later on is also unknown. In addition, the mortality or morbidity risk of antepartum hemorrhage in the absence of medical care has not been determined. Finally, data are scarce with regard to the impact of interventions targeting antepartum hemorrhage. In developed countries, management of antepartum hemorrhage is frequently determined on a case-by-case basis since its etiology varies and management is dependent on multiple factors including etiology, the status of the mother and fetus, the amount of bleeding, gestational age, and in the case of placenta previa and abruption, the degree of separation between the uterus and the placenta. In developed countries, where comprehensive maternal care is not only high quality but also widely and promptly available, the mortality risk of antepartum hemorrhage has been reduced to <1%. This low mortality risk is attributable to a highly vigilant approach to this condition, generally consisting of: (1) emergency caesarean section for patients with refractory hemorrhage, poor fetal status, or significant bleeding after 34 weeks gestation; (2) hospitalization with close monitoring and supportive care for actively bleeding patients; (3) expectant management as an inpatient (or outpatient if the patient lives within 5-10 minutes of a comprehensive medical center) with close follow-up and planned caesarean section (or vaginal delivery, if possible) at 36 weeks (after documentation of fetal lung maturity) or sooner, if necessary, for patients with a resolved episode of antepartum hemorrhage due to placenta previa or abruption. The level and intensity of care required is not feasible for most developing countries. Additionally, there are currently no established guidelines or effectiveness data concerning the management of antepartum hemorrhage using a less vigilant approach in resource-poor settings.

Incidence and case fatality rate
16 Globally, puerperal sepsis and infection are estimated to contribute to nearly 10% of all maternal deaths in Africa (9.7%), Asia (11.6%), and Latin America and the Caribbean (7.7%) [Khan 2006]. In India, puerperal infection and sepsis are responsible for 11% (range, 9-14%) of all maternal deaths [Registrar General 2006]. Initial estimates for the overall incidence of puerperal sepsis were created using data from the 2002 edition of the WHO's Global Burden of Disease study specifically for the SEAR region. We calculated estimates for the observed incidence of puerperal sepsis (0.041588) by using data on the number of cases (n=1,573,000) [WHO 2007a] and data on the number of births (n =37,820,000) [UNICEF 2004]. We base our estimates for the risk of puerperal sepsis on the 2000 GBD estimates [Dolea 2003b], that births occurring inside facilities with SBA were assumed to have a risk of puerperal sepsis of 2.5%. We assumed skilled birth attendants adhere to clean delivery practices, and therefore home deliveries attended by SBA had the same risk of puerperal sepsis. Those delivering at home with an untrained attendant had double the risk, at 5.0% [Dolea 2003b]. To account for the uncertainty in our initial estimates, we established a range of 4.2% -6% for sensitivity analysis.
A 2004 Cochrane Review that assessed the effectiveness and safety of antibiotic prophylaxis in reducing infectious puerperal morbidities in women undergoing operative vaginal deliveries included only one trial in which women underwent vacuum or forceps deliveries. While there was no statistically significant difference in the group of women that were given antibiotics versus those not given antibiotics, there was a relative risk reduction of 93% in the prophylactic antibiotic group [Liabsuetrakul 2004]. Two studies by Mosha et al. (2000) and Winani et al. (2007) concluded that women who bathed before delivery and women who used a clean delivery kit were 2.6 and 3.2 times less likely to develop puerperal sepsis than women who did not, respectively. Other studies reported a non-significant difference or inconclusive difference in effectiveness of interventions to prevent puerperal sepsis [Goodburn 2000;Hussein 2004;Bakr 2005;Tsu 2009].
The initial estimate for the case fatality rate (CFR) was from a review conducted by the Disease Control Priorities Project (DCP2) ] who reported an average CFR of 1.3% and CFR 3.9% for severe sepsis; these estimates were then adjusted according to heterogeneity in severity and underlying morbidity (e.g., severe anemia) by calibrating the model to fit multiple epidemiologic targets simultaneously. See section above on Adjustment of case fatality rates for heterogeneity in severity and co-morbidity. The implications of our adjusted CFR widened the implied plausible range. While the literature based range based on 0.013 average CFR from DCP2 was 0.009 -0.017, the expanded plausible range based on our adjustment of a CFR of 0.028 was 0.009 -0.036. Model projected mortality due to sepsis, as well as MMR, TFR, and calendar deaths for 2005, closely approximated the empiric data.

Reduction in mortality
We assumed the treatment regimen for puerperal sepsis (e.g., 2-day intravenous course of ampicillin, gentamycin, and metronidazole followed by an 8-day course of intramuscular gentamycin and oral metronidazole) had an overall treatment efficacy of 90% [Adam 2005, French 2003. A similar estimate was used in a recently published modeling analysis conducted by Pagel et al. (2009). Assuming an 11% case fatality rate for sepsis following delivery in sub-Saharan Africa, an 8-fold higher case fatality rate for sepsis without antibiotics compared to with antibiotics, and a 40% rate of antibiotic use, they estimated an 87.6% reduction in mortality from sepsis [Pagel 2009]. We varied this estimate from 63% to 93% for effectiveness expected in an appropriate EmOC facility.

Obstructed labor
Incidence and case fatality rate

17
The two major causes of obstructed labor are cephalopelvic disproportion and abnormal fetal presentation (i.e., breech or brow presentation). Major complications of obstructed labor include endometritis, rectovaginal or vesicovaginal fistula, and ruptured uterus with consequent hemorrhage, shock or death. If the obstruction cannot be resolved by manipulation (to reposition the fetus) or instrumentation (with forceps or vacuum to deliver the fetus), cesarean section is required. Globally, obstructed labor is estimated at 4.6% of live births, although varies considerably among different regions of the world [Khan 2006, Dolea 2003c. Approximately 5% (range 3-6%) of all maternal deaths in India are due to obstructed labor [Registrar General 2006], with the major cause being cephalopelvic disproportion. Women who are malnourished, marry young, or engage in childbearing at an early age before the pelvis has reached adult proportions, are at high risk for obstructed labor. However, there is no evidence to show that interventions aimed at providing adequate childhood nutrition or delayed childbearing prevents obstructed labor [Tsu 2009]. Using 2002 GBD data from the SEAR region, we estimated the incidence of obstructed labor (0.046822) by using data on the number of cases (n=1771) [WHO 2007a] and data on the total number of births (n =37820) [UNICEF 2004]. To account for the uncertainty in our initial estimates, we established a range of 3% -7% for sensitivity analysis.
The initial estimate for the case fatality rate (CFR) was from a review conducted by the Disease Control Priorities Project (DCP2) ] who reported an average CFR of 0.7%; this estimate was adjusted according to heterogeneity in severity and underlying morbidity (e.g., severe anemia) by calibrating the model to fit multiple epidemiologic targets simultaneously. See section above on Adjustment of case fatality rates for heterogeneity in severity and co-morbidity. The implications of our adjusted CFR widened the implied plausible range. While the literature based range based on 0.007 average CFR from DCP2 was 0.005 -0.009, the expanded plausible range based on our adjustment of a CFR of 0.019 was 0.005 -0.025. Model projected mortality due to obstructed labor, as well as MMR, TFR, and calendar deaths for 2005, closely approximated the empiric data.

Reduction in mortality
We identified multiple studies, including two Cochrane reviews that examined that efficacy of treating obstructed labor in reducing of maternal mortality rates. A study by Yarrow et al. (2004) showed a 94.1% success rate when using vacuum-assisted deliveries. Of the nine failed vacuum deliveries, four were subsequently delivered by forceps and five by cesarean section, with no maternal mortality reported. We assumed a 95% reduction in maternal mortality when obstructed labor was managed in an appropriate facility (assisted vaginal delivery with forceps or vacuum and, if necessary, cesarean section) [Adam 2005, Johanson 2000, Hofmeyr 2000,2003Schuitemaker 1997]. To account for the uncertainty in our initial estimates, we established a range of 76% -100% for sensitivity analysis.

Incidence and case fatality rate
Hypertensive disorders of pregnancy refer to a range of conditions associated with high blood pressure, proteinuria and, rarely, seizures. Severe pre-eclampsia and eclampsia have the highest case fatality rates of the hypertensive disorders of pregnancy, and can lead to placental abruption, disseminated intravascular coagulopathy (DIC), adult respiratory distress syndrome (ARDS), cerebral hemorrhage, seizures, and death. Globally, the incidence of pre-eclampsia is estimated at 3.2% of live births and eclampsia at 0.5% [AbouZahr 2004]. While in some parts of the world, such as Latin America and the Caribbean, hypertensive disorders of pregnancy are the leading causes of maternal deaths (25.7% of all maternal deaths) [Khan 2006], in India, hypertensive disorders of pregnancy rank as the fourth most common cause of maternal mortality [Dolea 2003d]. Eclampsia has a high case fatality rate, which varies among regions of the world, presumably as a function of the access to and quality of health care. [Dolea 2003d] A retrospective study of pre-eclampsia-and eclampsia-related deaths in Chandigahr, India found that access to and delay in seeking care was a major determinant of mortality, with 37.7% in grade IV coma and 54% with recurrent convulsions prior to admission [Sawhney 2000]. The GBD 2000 reported an incidence of 0.028 and 0.008 for preeclampsia and eclampsia respectively, for the SEAR D region. [Dolea 2003d] Initial estimates for the overall incidence of hypertensive diseases of pregnancy were based on data from the 2002 edition of the WHO's Global Burden of Disease study specifically for the SEAR region. The estimate derived from this data for hypertensive diseases was higher (0.066) from the SEAR region, [WHO 2007a;UNICEF 2004] but is consistent when corrected for the proportion of hypertensive disorders that are preeclampsia and eclampsia. Given a pooled average for preeclampsia of 0.034, if 2.3% of preeclampsia in SEAR D is an approximation of eclampsia (~0.0078), then the implied incidence of hypertensive disorders is 0.069, approximating the 2002 GBD estimate.
To account for the uncertainty in our initial estimates, we established a plausible range based on our literature review, including studies in the Cochrane database. The only interventions shown to prevent pre-eclampsia are anti-platelet agents, primarily low dose aspirin, and calcium supplementation. While data from trials are insufficiently conclusive as to the optimal timing of delivery with pre-eclampsia, there is robust evidence that magnesium sulfate can prevent and control eclamptic seizures, and for pre-eclampsia, reduces the risk of eclampsia by more than 50%. [Duley 2009, Langer 2008, Tukur 2009].
The initial estimate for the case fatality rate (CFR) was from a review conducted by the Disease Control Priorities Project (DCP2) ] who reported an average CFR of 1.7%; this estimate was adjusted according to heterogeneity in severity and underlying morbidity (e.g., severe anemia) by calibrating the model to fit multiple epidemiologic targets simultaneously. See section above on Adjustment of case fatality rates for heterogeneity in severity and co-morbidity. The implications of our adjusted CFR widened the implied plausible range. While the literature based range based on 0.017 average CFR from DCP2 was 0.012 -0.022, the expanded plausible range based on our adjustment of a CFR of 0.021 was 0.012 -0.027. Model projected mortality due to hypertensive disorders, as well as MMR, TFR, and calendar deaths for 2005, closely approximated the empiric data.

Reduction in mortality
We assumed that severe pre-eclampsia and eclampsia required treatment with intravenous hydralazine and magnesium sulfate; in addition, approximately 10% of all cases were assumed to require emergent cesarean section [Cahuana-Hurtado 2004]. A Cochrane Review of magnesium sulphate and other anticonvulsants for women with pre-eclampsia compiled evidence from 6 trials of which the largest source of data was the Magpie Trial Collaborative Group. [Duley 2003] This review found a 59% reduction in risk of eclampsia in women with pre-eclampsia (RR 0.41) and a 46% reduction (RR 0.54) in the risk of dying in women with pre-eclampsia randomized to magnesium sulfate. [Duley 2003]. A review showed that magnesium sulphate was the better anticonvulsant choice when treating women with eclampsia, and substantial reduced the risk of further seizures when compared to diazepam [Duley 2009]. One goal of this study is whether induction of labor in women with pregnancy induced hypertension or pre-eclampsia at term reduce costs and improve quality of life as compared to expectant monitoring. Two studies looked only at mild pre-eclampsia and gestational diabetes, but not at cases of maternal or neonatal death or eclampsia [Koopmans 2007[Koopmans & 2009 A study by one collaborative group found that the use of magnesium sulphate for women with pre-eclampsia was associated with a 16% reduction in the risk of death or serious morbidity related to pre-eclampsia 2 to 3 years later [Magpie Trial 2007] We assume that aside from the use of magnesium sulphate, induction of labor could occur in facilities capable of basic and comprehensive emergency obstetric care for women who do not require emergency cesarean section. Thus we rely on the higher effect size from the Cochrane Review, although still perhaps a conservative estimate, for the reduction in the case fatality rate of severe pre-eclampsia. We assumed severely pre-eclamptic/eclamptic women who received treatment had a 59% reduction in disease-specific mortality compared to those without treatment [Adam 2005;Magpie Trial Collaborative Group 2002;Duley 2003;Crowther 2002]. A Cochrane Protocol for additional evaluation of interventions for treating pre-eclampsia and its complications has been submitted and will be useful for further updating of this estimate once the review has been published. [Duley 2009]. To account for the uncertainty in our initial estimates, and based on the literature review, we established a range of 45% -95% for sensitivity analysis.

Pre-eclampsia and eclampsia
Relative Risk (

Initial estimates for complication (GBD data)
Neurological sequelae 0.0008 Severe anemia a 0.090 Sheehan's syndrome 0.008 Infertility from sepsis b 0.086 Fistula c 0.022 a This estimate takes into account the rate of severe and moderate anemia, the overall incidence of PPH, and the overall incidence of severe anemia in pregnant women, during the postpartum period, and in the general reproductive age group. We assume that women with pre-existing moderate anemia contribute disproportionately to the subsequent severe anemia observed following PPH. b Represented by the risk of PID (0.40) multiplied by the risk of infertility (0.22) given PID, to yield the estimate of 0.086 c We assume 25% are treated in Uttar Pradesh, Rajasthan, and rural India but vary this from 25% to 75% in sensitivity analysis.

20
The World Health Organization (WHO) defines "unsafe abortion" as "the termination of an unintended pregnancy either by persons lacking the necessary skills or in an environment lacking the minimal medical standards, or both" [World Health Organization 1998]. In India, unsafe abortion is responsible for up to 10% of all maternal deaths [Registrar General 2006]. Most abortions occur in married women with more than 2 children and who lack access to short-or longterm contraception [Ganatra 2002[Ganatra ,2006Mills 2007]. In one review of admissions for unsafe abortion to a tertiary care center in North India [Jain 2004] over a 15-year period (1988-2002) unsafe abortion caused ~17% of maternal mortality in the hospital. Initial estimates for the probability of an elective abortion were drawn from a published study that reported an abortion ratio of 17% (defined as the number of abortions per known pregnancies, including miscarriages) and studies that allowed for an approximate estimation of the proportion of elective abortion that is illegal or unsafe in South-central Asia [Henshaw 1999;Sedgh 2007]. We used other published data on hospitalizations for abortion-related morbidity and mortality to check the face validity of estimates, and establish a plausible range for sensitivity analysis. [Jain 2004;Singh 2006;Mills 2007;Coyaji 2002] These estimates were varied from 10% to 22% (abortion ratio), and from 25% to 50% (proportion of safe abortions), respectively. , Berer 2004, AGI 2006] Calibration exercises leveraged the information on distribution of direct causes of maternal mortality, including unsafe abortion, to check the face validity of estimates. [Registrar General 2006;Khan 2006] We used Asia-specific estimates from the WHO, and assumed that illegal/unsafe abortion is associated with a mortality of 300 per 100,000 procedures.  We developed an initial estimate of infertility from illegal/unsafe abortion of 12%, derived from the GBD study, in which the number of cases of infertility arising from unsafe abortion projected for India was divided by the number of unsafe abortion procedures projected for India [Murray 1998]. This was refined using more recent data, and a plausible range was established for the risk of infertility using higher and lower risk estimates reported for other world regions by the WHO [Ahman 2006]. In addition, we assumed a proportion of safe and unsafe abortion, 2.8% and 14.7% respectively, was associated with post-abortion complications requiring hospitalization and incurring quality of life decrements and costs [Johns 2007, Singh 2006].
The Indian Parliament passed the Medical Termination of Pregnancy (MTP) Act in 1971, which stated that abortion can be performed under the following conditions: "to save the woman's life, to preserve physical health, to preserve mental health, rape or incest, fetal impairment, economic or social reasons, and contraceptive failure." [Mills 2007] However, in most states services have been available in less than one-fifth of primary health care centres [Ramachandar 2002[Ramachandar , 2004[Ramachandar , 2005]. There are regional as well as rural-urban disparities in access to abortion services. [Ganatra 2006;Ipas, 2008] Barriers include untrained providers (and lack of availability and acceptability of trained non-physician providers), lack of equipment, cultural stigma, and lack of knowledge in women. [Hirve 2004, Shah 2005, Ganatra 2002] To provide broader access, a recent amendment to the original law in India decentralizes the approval of locations designated as MTP Centers from the state to the district level. As part of the Reproductive and Child Health program (RCH II), included within the National Rural Health Mission (NRHM 2005(NRHM -2012 there is a commitment to expand MTP facilities to make safe abortion services accessible to all women, particularly to those in the rural areas.

21
To estimate the risk of mortality from safe abortion the majority of our estimates were drawn from U.S. data in the early 1970s, a period in which elective first-trimester abortion was legalized in most U.S. states. Data from the Joint Program for the Study of Abortion (JPSA) demonstrate manual vacuum aspiration (MVA) is associated with fewer total and major complications compared with dilation and curettage (D&C). [Tietze 1971[Tietze ,1972, consistent with results from other US large, prospective studies [Edelman 1974] and smaller scale studies in developing countries [Laufe 1977]. By synthesizing British data for 439,400 legal, first-trimester abortions performed between 1968-1973 and U.S. abortion surveillance data from 1972-1975, we estimated the mortality risk for D&C was 1.8 per 100,000 procedures and MVA was 1.3 per 100,000 procedures [IOM 1975, Koonin 1993]. Our estimated mortality risk associated with MVA falls within the range of values (1.0 to 1.6 per 100,000 procedure) reported in published literature from the 1970s [Tietze 1971[Tietze ,1972IOM 1975]; however, we acknowledge the actual risk may be higher in developing countries ].
For sensitivity analyses on the complications and costs of medical safe abortion, we used data on vaginal misoprostol [Faundes 2007, Shannon 2004. We conservatively assumed a success rate of 80% for the vaginal misoprostol regimen, which falls roughly midway between reported estimates [Faundes et al. [2007] reported a success rate of 65%-93% for regimens using 800 mcg vaginal misoprostol (1 to 5 doses); Carbonell et al. [1998] found a success rate of >90% with 2400 mcg of misoprostol or in very early pregnancies terminated up to 9 weeks gestational age, compared to <90% in those terminated in the late first-trimester.] Major complications resulting from medical abortion include pelvic infection and hemorrhage necessitating transfusion and were estimated to arise in 0.75% of all procedures [Faundes 2007, Shannon 2004Grimes 2005;Reeves 2006].

Subsection B Data: Coverage Inputs and Selected Services
Coverage rates of skilled birth attendants and traditional birth assistants as well as of facility-based births were derived from national databases and published literature. [IIPS 2007 chapter 8] Using data from NFHS-3, more than half of births occur at home, as displayed below. [IIPS 2007] By birthing assistant, just under half of births are delivered by a skilled provider, with a major difference between states, and between rural and urban areas [IIPS 2007]. In the model delivery setting is differentiated by site including (1) home; (2) birthing center or health centre (used interchangeable here), (3) facility with bEmOC, (4) facility with cEmOC, and differentiated by health provider including (1) family member, (2) traditional birth attendant [TBA], (3) skilled birth attendant. Facilities classified as birthing centres or health centres are assumed to be staffed by SBA with expectant management of labor but do not have all signal functions to qualify as bEmOC. Extrapolating from data on the availability of emergency obstetric services and the relative availability of bEmOC to cEmOC, we estimated for the status quo, approximately 30% of facilitybased births occur in an EmOC-capable site, of which 17% offer cEmOC [AMDD 2002, IIPS 2007. We therefore assume in the base case analysis that for births that occur in a facility with EmOC capacity, approximately 90% are assumed to occur in bEmOC facilities and 10% in cEmOC facilities. In strategies that shift home births to facilities, additional analyses are conducted using several alternative assumptions. For example, as shown below, we explore in a scenario analysis the impact of (a) changing the distribution of routine deliveries that occur in primary facilities lacking EmOC (birthing centres/health centres) and facilities with EmOC, and (b) changing the distribution of deliveries in EmOC that occur in bEmOC versus cEmOC. In this particular analysis we also assume SBA-administered misoprostol in birthing centres/health centres.

Antenatal care
Data on antenatal care (ANC) from the NFHS-3 were used for coverage rates for the national model, and were stratified by urban and rural status when available. State-specific data were used for Uttar Pradesh, and were also stratified by rural and urban status when possible. [IIPS 2007 Chapter 8] We assumed in our analyses that antenatal care includes 4 visits, tetanus vaccination, syphilis, gonorrhea, chlamydia screening (and treatment), urinalysis, blood tests, treatment for anemia, counseling (e.g., family planning, spacing, intrapartum care).
National and state-specific ANC and anemia treatment rates (NFHS-3) [ In addition, Ram and Singh [2006], based on data from the District Level Household Surveys, found that utilizing antenatal care services may lead to the utilization of other maternal health services such as institutional delivery, delivery with skilled attendance, and advice-seeking behavior for pregnancy-related complications and postpartum complications. Sensitivity analyses were conducted to explore the range of potential benefits associated with the scenarios that antenatal care increases facility over home births and the use of emergency care for those remaining at home.

Family planning
Our choice of including a comprehensive strategy of enhanced family planning to reduce the unmet need for contraception, for purposes of both limiting and spacing, reflects the overarching goal of the Government of India to address the unmet need for contraception and bring the total fertility rate down to replacement level by 2010 (IIPS 2007). The effectiveness of family planning is incorporated into the model as a set of variables that reflect (i) Coverage level of contraceptive method; (ii) Distribution of contraceptive type; (iii) Type-specific failure rate. We use state-level and setting-specific data to represent the current met need for contraception and the distribution of methods used by age. Failure rates are conditional on the method used.  (IIPS 2007, ch 5, p.160). In our base case analysis both age-patterns and rural/urban status are incorporated, but we do not stratify according to education and wealth. We do explore age-specific focused interventions in sensitivity analyses, specifically, focusing efforts to increase modern contraception in younger women. The motivation for these exploratory analyses was based on data showing that although over 40% of women use modern contraception, family planning is used mainly for the purpose of providing long-term contraception in India (IIPS 2007). This is reflected in the low contraceptive prevalence in younger females and the overwhelming domination of the contraceptive method mix by sterilization. For example, only 5% of married women ages 15-19 years use modern contraception versus 67% of women ages 35-39 years. Female sterilization is the most common method of contraception, accounting for more than 75% of total contraceptive use. Women below age 25 are more likely to have used modern and traditional spacing methods, whereas women age 25 and over are more likely to have undergone sterilization. In fact the median age for women undergoing sterilization is 26 years, illustrating the typical childbearing pattern of women in India  Contraception prevalence rate is 56 percent (IIPS 2007, ch5, p.120) and higher in urban than rural areas. Overall, 86% of those using contraception use modern methods and 14% use traditional methods. Female sterilization accounts for 77% of modern methods used, and the prevalence is similar in urban and rural women. Condoms and rhythm method are used commonly for spacing, and modern spacing methods (pill, IUD, condom) are higher in urban areas compared with rural areas. Condom use is 3-fold higher in urban areas. Among sexually active unmarried women age 15-49, 36 percent report using a modern method. Women younger than age 25 are more likely to have used modern and traditional spacing methods, whereas women age 25 and over were more likely to have undergone sterilization.

Overview of our strategies to increase contraceptive use
We elected to use unmet need as our main intervention target to describe the influence of increasing access and uptake of modern contraception, as it reflects two groups of women: (a) women who are not using any method of contraception but who do not want any more childrenunmet need for limiting and (b) those who are not using contraception but want to wait two or more years before having another child -unmet need for spacing. [The sum of the unmet need for limiting and the unmet need for spacing is the modeled unmet need for family planning]. For India as a whole, there has been a decrease in the unmet need for family planning from 16 percent in NFHS-2 to 13 percent in NFHS-3. The decrease in the unmet need for spacing was higher than the decrease in the unmet need for limiting. NFHS-3 showed that in most states the unmet need for limiting is higher than that for spacing. The table below illustrates differences in baseline model assumptions about access to family planning (e.g., the unmet need for spacing and limiting births) and the magnitude of stepwise increases characterizing different strategies.  (1) knowledge about contraception; (2) ever use of contraception (measure of the cumulative experience of a population with family planning); (3) stage of family-building at time of contraceptive uptake and method choice; (4) intention to use a method of contraception in the future; (5) source of contraception by type (e.g., public sector, private sector) and location of sterilization (e.g., government/municipal hospitals, community health centers). If data on interventions that leverage this information becomes available, it could be integrated in future analyses.

Relationship between abortion and contraception
The risk of unsafe abortion is reduced through the use of contraception, legalisation of elective abortion, and the use of safe abortion methods by a high-quality and trained provider. [AGI 2007] Access to safe, effective contraception can substantially reduce the need for abortion to regulate fertility. [Mills 2007] Mills et al. [2007 report 64.9% of women in Uttar Pradesh sought elective abortion because of an unwanted pregnancy whereas less than 15% sought abortion because of complications during pregnancy or illness. We therefore explored correlations of 25% to 75% in the model. The model is used to generate the reduction in unsafe-abortion related deaths due to increased access to and use of family planning (modern methods), in addition to the averted deaths due to safer abortion and postabortion services.

Postpartum care
In the recent NFHS-3, women reported complications two months after their most recent deliveries, including massive vaginal bleeding for 12% of births and a very high fever for 14% of births; both complications were more common among rural than urban mothers. [IIPS 2007 chapter 8] Bang et al. [2004] document the high incidence of maternal morbidity during labour and puerperium in rural homes in Gadchiroli, India, with more than 40% of women experiencing postpartum morbidity. While it was difficult to find quantitative data on reduction in mortality with postpartum care, we did conduct sensitivity analysis to estimate the potential averted morbidity and costs. In addition, we explored the potential benefits associated with increased use of contraception and other reproductive health services associated with postpartum care. The Lancet series on maternal survival suggested that the evidence in support of communitybased interventions, especially those geared towards reducing maternal mortality (such as the use of misoprostol by TBAs to reduce PPH in the home setting or clean birth kits in reducing death from sepsis in the home), is weak. [Campbell 2006]. A Cochrane Review of the evidence on the effects of TBA training for improving health behaviors and pregnancy outcomes found insufficient data to document any association between training and maternal mortality. [Sibley 2007] Through a review of the recent literature, we identified several trials that investigated the use of community-based approaches to improve maternal health, most involving the use of misoprostol to reduce mortality from PPH. Sanghvi et al. (2004) reported that community volunteers were able to successfully distribute oral misoprostol tablets to women and encourage the acceptance and use of the tablets; women in the intervention area were 45% less likely to need an emergency referral for PPH. Prata et al. (2005) also found positive results from training TBAs to diagnose and treat PPH with misoprostol, demonstrated through lower referral rates for women with PPH in the intervention area than for women in the control area (2% versus 19%). In an extensive review of the literature on interventions suitable for resource-poor settings, Prata et al. (2009) conclude that even TBAs are able to provide some basic maternal care including misoprostol for PPH. Walraven et al. (2005) investigated the use of misoprostol versus placebo for management of the third stage of labor for home births under guidance of trained TBAs but found a non-significant reduction in severe PPH.

Maternal care indicators by state, India (NFHS-3) [IIPS
In rural India, a randomized controlled trial of the use of oral misoprostol to prevent PPH in a home birth setting found an 80% reduction in the rate of acute severe PPH in women given misoprostol as opposed to expectant management of the third stage of labor [Derman 2006]. Recognizing that the evidence is still quite limited, a Cochrane Protocol has recently been submitted to assess the effectiveness of community-based intervention packages for preventing maternal mortality and morbidity. [Haider 2009] An analysis in which we simulate the community-based interventions included in Prata et al. (2009) has been included in the Results Section of this document.

Coverage of community-based interventions in India
Aside from the need for convincing evidence on the effectiveness of community-based interventions to improve maternal health, additional evidence on the potential reach of such interventions across a large portion of the population is required to justify policy relevance of these strategies. Prior studies in India have found conflicting results on the ability to achieve high coverage levels of community-based interventions. [Bang 2005a, 2005b, Baqui 2008, Patel 2010] Bang et al. [2005aBang et al. [ , 2005b achieved high coverage in using trained village health workers (VHWs) to provide home-based neonatal care in a rural part of India; home-based care was provided to 93% of neonates. [Bang 2005a] Encouraging VHWs to be present during home delivery with a small ($1.00) financial incentive resulted in VHWs attending 84% of home deliveries. [Bang 2005b] However, most community-based trials are conducted under controlled conditions, thereby ensuring high program coverage. [Baqui 2008]; Methods to achieve scale-up are needed. [Bang 2005a] Baqui et al. (2008) evaluated the effect of a community-based package of maternal and newborn interventions implemented at scale using existing government infrastructure. While community-based workers were present throughout the study area, 38% of women who had recently delivered had not received any home visits (antenatal or postnatal). Patel et al. (2010) concluded that while there is potential for community health workers to deliver interventions on a large scale, their effectiveness will be limited by a weak rural health system, lack of incentives, absenteeism, and lack of support and supervision. Singh et al. (2009) investigated the potential of a public-private partnership scheme in which the government of Gujurat paid private obstetricians practicing in rural areas to provide free delivery care to poor women. Out of the payment per delivery, obstetricians were required to pay the woman giving birth for transport to reduce the delay in seeking delivery care, and the person who accompanied the woman. This scheme was found to be very successful during the trial period, in that numerous obstetricians have joined the scheme, and the estimated coverage of deliveries among poor women increased from 27% to 53%. ]

Barriers to Effective Referral to EmOC
Effective referral relies on the ability to overcome three critical delays (a) recognition of referral need and willingness to be referred (by provider and delivery location); (b) expedient transfer to referral facility (determined by distance, affordability, available transport); and (c) timely treatment in an appropriate facility capable of high-quality emergency obstetrical care (e.g., 6 signal functions in bEmOC, blood transfusion and surgery in cEmOC). We expanded Thaddeus and Maine's [1994] "three delays" framework to reflect the multidimensional nature of each of these delays and the heterogeneity between and within countries as to which delays and components are most critical. A successful referral in our model incorporates a series of elements, each of which could act as a barrier to the care a woman with pregnancy-related complications requires.

Delay Category 1. Recognition of need for referral and/or willingness to be referred
We include in this category both failure or delay in recognition of the need for referral by the SBA as well as delay in recognition for need of referral or unwillingness to be referred on the part of the woman or her family. We assumed the recognition rate for complications developing during home deliveries would vary based on the level and skill of the birth attendant. Based on data from Honduras regarding traditional (untrained) birth attendants, [Danel 2003] we assumed an 11.5% recognition rate for unskilled delivery at home, and a 20% recognition rate for skilled birth attendants at home.  We assumed life-threatening complications (those needing cEmOC capability) occurring at bEmOC were recognized as needing transfer to a facility with cEmOC. We also included an analysis assessing the impact of delays in facility transfers (i.e., incorporating the delay due to transport problems, logistics, or fees). In addition, we assumed an "erroneous" referral rate (in the absence of complications), owing to misdiagnosis and lack of patient monitoring support, that varied from 2.5% to 10% based on location of delivery and skill level of birth attendant.

29
A literature review on the persistence of poor maternal health in Rajasthan [Iyengar 2009a] revealed that although 73% of women had contacts with health professionals during pregnancy, less than one-sixth of women received advice about danger signs or place of delivery. Gupta et al. [2006] reported on maternal mortality in Uttar Pradesh, finding that almost 2/3 of the 286 maternal deaths occurred before reaching the health facility; about half occurred at home and another 12.6% of the cases died during transit. More than one-third of women were delayed for referral because they were not considered serious enough; an additional 32% were delayed at home because the woman refused referral (10%), money could not be arranged (16%) or there was no one to take care of family at home (6%). Mills et al. [2007] reported in an analysis of 45 maternal deaths in Uttar Pradesh with delays in obtaining appropriate care, that the decision to seek care took more than a day in nearly half of the cases. In some instances the initial decision was to first seek care from a non-health professional; by the time a woman finally reached the appropriate health facility, it was too late or the woman ended up dying en route. Finally, in Andhra Pradesh, Prakasamma [2009] found that women did not go to primary health centers (PHCs) for childbirth since there were no service providers or facilities; results of community studies indicated the unwillingness of people to use these facilities for maternal emergencies. Of the 148 deaths in the study, 20.3% died on way to the hospital, and 13.5% died at home, indicating a need for earlier recognition and willingness to be referred. The Tables below show the initial range of baseline estimates used in sensitivity analyses, and the initial range across which stepwise improvements were made in the temporal strategies evaluated. These ranges were expanded in a series of exploratory analyses. We made assumptions about effective transfer that varied by both delivery location and urban/rural setting, and were intended to reflect access to transport, reliable fuel and accompanying person en route, and interim care if necessary. [Government of India 2008aIIPS 2005Padmanaban 2009;Iyengar 2009a] This ranged from a high of 81% in urban India to a low of 24.4% when deliveries began at home in rural India. We assigned 81% (and not 100%) as the baseline rate of "expedient accurate referral" in an urban setting lower-level facility, to reflect delays attributable to multiple transfers between facilities, and delays related to being turned away from one hospital and having to travel to another,.

30
Based on other literature, and India-specific studies, we established a plausible range for sensitivity analysis. A study in Uttar Pradesh, Karnataka, Uttranchal, Maharashtra and Delhi reported 55.6% of women did not seek treatment because of a transportation barrier. In the maternal deaths evaluated, approximately 25% occurred at home in the absence of referral and 15% en route. [Pandey 2003] Studies in the Indian states of Andhra Pradesh, Maharashtra, [Ganatra 1998] and Rajasthan found that 42% to 52% of maternal deaths occurred at home or in transit to a hospital. [Mavalankar 2005] In a qualitative assessment in Uttar Pradesh, [Mills 2007] nearly 19% of deaths occurred en route to a health facility. A survey of maternal deaths from seven districts in Uttar Pradesh found that nearly one quarter of families were unable to arrange transport, and 16% did not have enough money to pay for the transport, summing to 41% [Gupta 2006]. According to this study, while 50% of maternal deaths occurred at home, 12.6% of women died while in transit to a facility.
A study in 3 states in India [Murthy 2004] concluded that ~62% of deaths and 41% of cases with complications experienced multiple referrals (usually first to a primary health center, which was inappropriate for 60% of cases resulting in deaths; primary health centers then referred women further leading to a loss of 3-4 hours of critical time). Distances to referral locations were much greater in cases of deaths, and took significantly longer to reach these locations (18% reached within 1 hour  Iyengar et al. [2009a] report on referral and transport from home and from facilities, finding that despite financial incentives and assistance for transport, often women did not use these funds. Among the reasons provided were that a functional vehicle and telephone were not available in half of the facilities, severely limiting the ability to provide prompt referral. The Tables below show the initial range of baseline estimates used in sensitivity analyses, and the initial range across which stepwise improvements were made in the temporal strategies evaluated. These ranges were expanded in a series of exploratory analyses. Facilities with basic EmOC (bEmOC) are assumed to be capable of administering injectable antibiotics, oxytocics, and sedatives or anti-convulsants, performing manual removal of placenta, removal of retained products, and assisted vaginal delivery. Facilities with comprehensive EmOC (cEmOC) also are able to provide blood transfusion, cesarean section, and management of advanced shock.

Range of baseline assumptions for status quo used in sensitivity analysis
We recognize that some tertiary sites will not have a blood bank and some secondary sites may eventually be able to perform c-section; further, we recognize that in the strategies that include stepwise investments in infrastructure and facility improvements, not all facilities will be expected to be fully implemented as one of the three distinct types.  [Ma 2008] found substantial shortfalls in India at each level of health facility, e.g., 50% fewer community health centers (CHCs) than needed [Datar 2007]. Moreover, care provided by the public sector is repeatedly described as poor, and care provided by the private sector is often deemed uneven [Bhatia 2004, Mills 2002]. In Uttar Pradesh, of the 15 district hospitals surveyed, 10 provided cEmOC (67%), 1 provided only bEmOC, and 4 did not qualify as either (27%). Of the 54 first referral units (FRUs), 1 qualified as cEmOC (1.9%), while 43 did not qualify as bEmOC (10 were bEmOC, 18.5%). Of all facilities (n=69), 15.9% were cEmOC, 14.5% were bEmOC and 70% did not meet the requirements to be either bEmOC or cEmOC [Mills 2007]. Only two of the 54 FRUs had blood banks in 2006.  Katrak [2008] reprted that rural and urban areas differed in the total number of medical practitioners, as well as in the types or 'quality' of the practitioners. Deshpande et al. [2004] found that rural areas in one district in India had a smaller number of qualified doctors and a larger number of unqualified practitioners. Katrak [2008] combined those categories to create an overall measure of the number of practitioners, finding that women in rural areas faced greater challenges in accessing medical practitioners than in urban areas [also noted in Perry 2000, Wagstaff 2002, Buor 2003, Leonard 2003, Dzator 2004. Because rural populations tend to be spread over a larger area, longer distances for travel are required to reach medical care and patients may have longer 'waiting times'. Katrak [2008] calculated a needs-to-access ratio, with the rural ratio in India at least 4.95 times greater than in urban areas and at most 14 times greater. An average of these two estimates implies that the rural shortage is about 9.5 times greater.

Subsection C Calibration Exercises and Model Performance
Calibration targets for each national and province-specific model are established based on survey data and published studies, and include the distribution of causes of maternal mortality (e.g., PPH, obstructed labor, sepsis), maternal mortality ratio (MMR), and the total fertility rate (TFR). The MMR is adjusted directly in the model for indirect causes of maternal-related mortality, as explained below. The importance of using multiple indicators is that different aspects of maternal mortality are reflected by each of them. For example, the MMR is not age-standardized, nor does it take into account the fact that women face the same risk numerous times over their reproductive lifespan, nor does it account for the reduction in risk attributable to declining fertility from family planning. The model can be used to project a range of maternal health indicators and these can be used as calibration targets, or can be compared to survey data to assess an approximation of face validity or projective validity. These include: Maternal mortality rate Defined as the number of maternal deaths per 1,000 women or 100,000 women of reproductive age (ages 15-45) or woman-years of risk exposure, and designed to be an indicator of risk of maternal death (i.e., cause-specific death rate) Proportionate mortality ratio Defined as the proportion of all female deaths among women of reproductive age due to maternal causes Lifetime risk of maternal death Reflects the probability of a maternal death during a woman's reproductive lifespan (the probability that a 15-year-old will eventually die from a maternal reason up to age 45, for example) and is described in terms of odds (it accounts for the probability of dying from maternal causes each time a woman experiences a pregnancy, and so takes into account fertility as well as obstetric risk) Lifetime risk of dying from maternal causes The calculation of lifetime risk assumes no changes in fertility or mortality; estimates are generated from the maternal mortality rate, and do take into account the competing causes of death. In contrast, in our model, the simulation over time does take into account the changes in fertility and background mortality, including changes in maternal mortality.

MMR
We used selected data for the India model MMR calibration target. We prioritized the recent estimate of 450 based on the reassessment of data by an international experts group that estimated the MMR to be 1.5 times the 2003 SRS estimate [Hill 2007;WHO 2007b]. We also took into consideration as an upper bound the estimate of 540 as reported in the 2005 World Health Report 2005, and as summarized by Mills et al. [2007]. Below, the range of MMRs using different sources and methods is provided. It is widely accepted that the error and uncertainty in these measures is formidable, and trends should be interpreted with grave caution. As sample sizes decrease, such at the state or district level, interpretation of trends should be avoided. That being said, the general pattern and rank ordering of state-level MMR estimates do provide an approximate categorization of states relative to one another. For example, there is a consistent rank order of urban India (best), India overall (next best) and rural India (worst), when considering the three main models and national and stratified data.

Revision of 2005 estimates of MMR for India
UNICEF, WHO, and the UN Population Fund (UNFPA) previously developed global, regional, and country estimates of maternal mortality for the years 1990, 1995. [WHO, UNICEF. 1996WHO, UNICEF, UNFPA, 2001;WHO, UNICEF, UNFPA, 2004] In 2006 a collaborative effort involving the WHO, UNICEF, UNFPA, the World Bank, and outside technical experts, reviewed the available data and developed revised estimates of maternal mortality for 2005. Countries were categorized on the basis of data availability, quality and type, and consensus was reached on methods of evaluation, data synthesis, and statistical modeling for estimation in countries with limited data. India was classified with China into a group characterized as estimating MMR from data from disease surveillance or sample registration systems. The recommendation for these countries was to consider the observed value as a lower uncertainty bound, double it for the upper bound, and multiply the observed value by 1.5 to yield the point estimate. [Hill 2007] We parameterized the state-level models for Rajasthan and Uttar Pradesh using the best information available and adjusted for the TFR as reported in NFHS 3. We then compared the model-projected MMR with reported data from those states. Note that the Rajasthan and Uttar Pradesh models were not calibrated to the MMR; rather we sought to assess model performance by comparing the model-projected MMR to reported data.  SRS 1999SRS -2001. [Registrar General 2006] The lifetime risk of maternal deaths was estimated to be 1 in 53 (1.9%). When the MMR was 627, based on older data [Bhat 1995] the proportion of maternal deaths of all deaths among all deaths of women of reproductive age was 29%.

Distribution of causes of maternal deaths -calibration data
A systematic review and analysis of the magnitude and causes of maternal deaths documents variation both across and within geographical regions. [Khan 2006] Estimates of specific causes of death in India are hindered by the same methodological challenges as in global estimates, further complicated by the considerable heterogeneity that exists. We used Khan 2006 regional estimates based on the large sample sizes, and took into consideration data from India as summarized below. The concordance was considerable in these estimates. Of note, anemia exerts a huge toll, contributing to 24% of all maternal deaths in one hospital study [Pendse 1999]. More recent data corroborate the role of anemia in maternal mortality (generally classified as an indirect cause) with Khan reporting 12.5% (Khan 2006) and Mills reporting 15% [Mills 2007].

39
The MMR is adjusted directly in the model for indirect causes of maternal-related mortality. We assume that the proportion of mortality that is categorized as indirect and attributable to anemia (~15%) will be reduced with strategies that include enhanced family planning, increases in appropriate antenatal care with completed courses of treatment for anemia, facility-based births with quality intrapartum care, and reliable access to basic and comprehensive EmOC. We conservatively assume that the proportion of mortality that is categorized as indirect and attributable to other causes will not be impacted.  b distribution of severe life threatening complications was adjusted according to basic and comprehensive EmOC need, and is described in section II of this document in the section "severity of complications"; an explanation of the distribution of causes is described in section 5 of this document.

Adjustments to initial estimates of variables in calibration exercises
Model performance was assessed by comparison of model-based projections with reported measures such as life expectancy, proportionate mortality ratio, and population-based outcomes [WHO 2006, Registrar General 2006, AMDD 2002, UNICEF WHO UNFPA 2007, UN 2007. Projective validity of the empirically-calibrated model was further assessed by simulating Rajasthan and Uttar Pradesh, and comparing projected maternal health indicators with reported data. [Registrar General 2006, Mills 2007 For the latter exercise, we parameterized the state-level models for Rajasthan and Uttar Pradesh using the best information available and adjusted for the TFR as reported in NFHS 3. We then compared the model-projected MMR with reported data from those states, e.g., the model projects a met need for EmOC, defined as the percentage of hospitalrequiring complications actually treated in a referral-level facility, of 9.4%, which was within the range of the met need of EmOC of 5.3%-12.2% observed in Rajasthan, India [AMDD 2002].

Overview: cost identification and measurement
Direct health care costs include the cost of a normal pregnancy (e.g., prenatal visits, normal labor and delivery), the cost of induced abortion, the cost of treating abortion-related complications, the cost of treating pregnancy-related complications (e.g., eclampsia, hemorrhage, sepsis), salaries of health care providers (e.g., counseling, skilled birth attendants, clinician time); costs related to prenatal care (e.g., additional prenatal visits, nutritional supplementation, treatment of anemia or other existing disease, screening for sexually-transmitted diseases [STDs]), providing safe abortion (e.g., manual vacuum aspiration) or family planning options (e.g., sterilization, intrauterine device [IUD], oral contraceptives), and emergency obstetric care (e.g., facilities with the capacity for transfusion, parental antibiotics, surgery, anesthesia). Direct non-health care costs include, but are not limited to, the costs of transportation to and from the clinic or provider, and costs of patient time seeking care or receiving care. Cost estimates are broken down by input (e.g., drugs, vaccines, salaries, infrastructure), by intervention (e.g., management of a normal birth, hemorrhage, eclampsia, sepsis), and by service location or level (e.g., hospital, health center, health post). Personnel cost (salaries) and facility costs are country-specific from International Labour Organization databases and data publicly available from the World Health Organization (WHO), and modified using state-specific data when available. Salaries originally reported in year 2000 International dollars were converted to local currency units using Purchasing Power Parity conversion rates, inflated to year 2006 local currency units using GDP deflators, and then converted to 2006 US dollars using exchange rates. When possible, we conducted literature reviews for costs associated with different services and complications; these costs were extrapolated and adjusted to the same year and currency to facilitate comparison and generate plausible ranges for each cost estimate. Differential costs of scale-up were assessed, as were training costs (e.g., for SBA).
Costs are presented in currency units that remove price inflation, and for analyses intended to inform resource allocation and compare studies from multiple countries, costs are expressed as US dollars or international dollars. While exchange rates may reflect under-or overvaluation of the local currency, they represent what is actually paid for locally produced inputs. Purchasing-power parity rates, in contrast, attempt to express what the local currency is worth in purchasing power, and therefore account for differences in price levels across countries. The exchange rate for domestic currency into international dollars is the amount of domestic currency required to purchase the same quantity of goods and services as $1 could purchase in the US.

Documentation of costs used in the India model
The India model requires country-specific estimates of all maternal interventions including safe abortion and long-term complications. Estimates in the current model were drawn from the UNFPA's Reproductive Health Costing Tools Model (RHCTM) [UNFPA 2007]. This model is designed to help countries estimate the cost of scale up for a basic package of reproductive health services -ranging from family planning, antenatal and delivery care to emergency obstetric care and HIV/STI prevention and treatment.

RHCTM: Essential package of 45 reproductive health interventions
The second part costs out activities and investment required to improve the health system of a country in order to scale up and provide the above package of reproductive health interventions. This includes investments in the physical and human infrastructure (building, rehabilitating, and equipping medical facilities; training and retaining staff; improving the referral and medical supply system) as well as demand creation, outreach, supervision, monitoring and evaluation activities.
The RHCTM uses an ingredients approach to estimate the costs associated with an intervention. Each complication is associated with a drug, supplies, and personnel requirements for treatment. However, the estimate does not include costs associated with occupying a health facility bed or an outpatient visit; these costs were obtained from the WHO CHOICE database and India-specific estimates were used. The RHCTM does not include estimates of facility costs per case. For this, we drew on the Indiaspecific estimates of unit costs for patient services provided in WHO-CHOICE. Since many interventions can occur outside a 20-minute visit time frame (e.g., 5 minutes or 30 minutes), we broke down the cost for outpatient visits according to an estimated cost per minute. In the following intervention-specific sections, we present cost estimates used in the model, followed by tables outlining how these costs were derived from the RHCTM for the following intervention components: (1) drugs and supplies per case; (2) personnel costs per case; and (3) facility costs per case.

Oral Contraceptives
We assumed that oral contraceptives are distributed at a health post and the cost under the current standard of care in India is $10.64 per year.

Condoms
The cost of providing condoms is $8.40 per year. We assumed that condoms are obtained at an outpatient health post visit. In a scenario where the use of family planning is increased, costs could be reduced through alternative delivery methods that eliminate this visit and its associated costs. The cost of treatment for anemia is added to the cost of prenatal care shown above. The management of severe anemia is not specified in the RHCTM, but was adapted based on the PPH protocol of the RHCTM, $1.02. A sensitivity analysis was conducted to assess a community based intervention for SBAadministered misoprostol to reduce incidence of PPH in home deliveries and in birthing centres. We assumed $4.40 ($3.40 based on the upper bound used by Sutherland and Bishai (2009) of training costs for SBA, and .99 for misoprostol). These costs represent the incremental costs above routine SBA delivery.

Incomplete abortion
We assumed all women with incomplete abortion from miscarriage were managed with manual vacuum aspiration at a cost of $8.90.

Elective abortion
Available data on the cost of abortion in India are sparse and do not distinguish between the cost of safe and unsafe abortion. We assumed the cost of elective abortion was 750 Rupees (Rs) based on a published study with a range of 500-1000 Rs [Duggal 2004]. This estimate was inflated from year 2000 Rs to 2006 Rs and then to 2006 US$, for a total cost of $21.87.

Supplemental literature-based estimates for the cost of elective abortion
The cost of safe and unsafe abortion was also estimated in a series of unpublished studies available from CEHAT. The year associated with these costs is not reported but since the CEHAT studies commenced in the year 2000, we assume the costs reported are in year 2000 Rs. Of note, CEHAT also provides an estimate of the out-of-pocket cost of spontaneous abortion ( . These reports were used to establish plausible ranges for an upper and lower bound for sensitivity analyses.

47
We assume that for births that take place at home there are three possible levels of care: assistance by a family member, assistance by a traditional birth attendant (TBA), and assistance by a skilled birth attendant (SBA). For home deliveries, we include the cost of the attendant's time but there are no facility charges, and in the base case, no charges for drugs (except for in the case of special analyses, in which we assessed the use of misoprostol). Deliveries at a primary-level health post (i.e., subcentre or primary health centre) utilize the following assumptions: all deliveries are attended by skilled staff, requires half a bed day, has mechanisms for referral to facility with EmOC; functions of bEmOC are not assumed to be present. For deliveries at a secondary-level health centre (i.e., community health center or first referral unit), basic EmOC is expected to be available. For deliveries at a tertiary-level facility (i.e., district hospital or medical centre), comprehensive EmOC is expected to be available. To model the relationship between the costs of delivery estimated using the RHCTM tool ingredients-based approach for normal delivery and the costs of delivery at different levels of facilities, we applied a scaled factor to reflect the higher costs in a tertiary hospital versus secondary-level facility versus a birthing centre or primary care health post. The scaling factor relied on the relative costs reported in the regional and country-specific WHO CHOICE databases for primary-, secondary-and tertiary-level facility bed-days and visits, as well as the relative costs of health provider salaries based on a distribution of increasingly skilled health providers comparing a primary-level birthing centre or subcentre to the tertiary-level hospital (e.g., specialists versus medical officers versus nurses). Results are shown below with bEmOC facilities being 1.7 times (1.30 -1.97) more costly than primary-level health centres, posts or birthing centres; tertiary facilities such as district hospitals and general cEmOC capability are 2.25 (1.78-2.56) times more costly than primary-level facilities. Mills et al. [2007] reported costs in Uttar Pradesh approximately 3.2 times higher in the district hospital compared with a community health centre; this was used as an upper bound in sensitivity analysis. Data from rural Rajasthan reported similar results [Iyengar 2009a] with delivery costs at government district hospitals approximately 2.47 times higher than a government primary health centre. All costs were varied in sensitivity analysis. (2) secondary facilities with bEmOC capacity; and (3) tertiary facilities with cEmOC capacity. We recognize that some tertiary sites will not have a blood bank and some secondary sites may eventually be able to perform c-section; further, we recognize that in the strategies that include stepwise investments in infrastructure and facility improvements, not all facilities will be expected to be fully implemented as one of the three distinct types. However, because the costs, functions and staffing are fairly closely aligned with basic or comprehensive EmOC capacity, this simple categorization captured the most important dimensions for purpose of this analysis. b

Facilities in the model
Public-health facilities in India as described in   To place our base case estimates in context, other delivery cost sources are shown below. In Uttar Pradesh, delivery fees were found to be much higher in private nursing homes than in private hospitals. [Mills 2007] The average fee for a normal delivery in a private nursing home was almost 2,000 Rupees (US$44, at 45 Rupees to the dollar), 2.5x the cost in a private hospital, 7x the cost in a district hospital. A sample of recent delivery costs reported for rural Rajasthan [Iyengar 2009a] and poor women in Gujarat ] are shown below. A sample of older delivery costs, reported by Borghi et al. [2006], is also shown below.

Emergency/pre-referral care (transportation costs)
Costs for transportation include the cost of transportation for a woman with a recognized complication that cannot be treated at the original birthing location (either true complication or false referral), and the cost for an attendant to accompany the woman during transport in some circumstances. Of note, the model permits additional costs to be assigned to the category of transfer from home to a facility that reflect interim lifesaving measures such as management of PPH and use of the non-pnuematic anti-shock garment (NASG), although this is not included in the base case for this analysis.

49
There is a relative lack of data available on the cost of transportation in India from one birthing location (home, health post, bEmOC), to a higher-level facility (health post, bEmOC, cEmOC). We were able to leverage some data on transportation from recently published papers Iyengar 2009a], and used data from other sectors to approximate increased costs associated with improving transport programs in rural regions.
2009 ;Iyengar 2009a], and used data from other sectors to approximate increased costs associated with improving transport programs in rural regions.

Get Distance and Cost Get Distance and Cost
Using public access data, we also developed a transport cost calculator that incorporates information on distance, road density, cost of vehicle and fuel, and approximates a cost. Data used to develop this cost estimation tool were derived from various sources including the World Bank and national databases on road transport. [World Bank Data Pages, IEA 2006, Automobile India, Government of India 2008b This allowed for rough approximations for face validity using studies that reported mean distance. For example, Bhat et al. [2009] reported for Gujarat state that the mean distance traveled for women participating in a public-private partnership (Chiranjeevi Scheme, which focused on increasing institutional delivery and emergency obstetric care for the poor) was 13.8 km but ranged from 1 to 72 km. Using the tool, we estimated the mean cost to be $2.90 with an upper bound of $8.00.

Management/treatment obstructed labor
Management of obstructed labor at bEmOC consists of assisted vaginal delivery with vacuum or forceps. In addition to obstructed labor costs, we also included the cost of prolonged labor, which precedes the diagnosis of obstructed labor. At the bEmOC level (where cesarean section is not available), the total cost is $19.51. In comparison, Bhat et al. [2009] reported a cost per procedure involving forceps, vacuum, breech (possible in bEmOC) of $22 in Gujarat. At the cEmOC level, the cost of c-section is $92.92. In comparison, Bhat et al. [2009] reported the cost per procedure involving c-section (as would be conducted in cEmOC) was $111 in Gujarat.

Management/treatment for postpartum hemorrhage
Managing PPH with basic services at a bEmOC-level facility costs $29.35; for those cases needing transfusion, advanced shock management, and/or surgery at cEmOC, the cost is $141.67. The difference in the cost of management/treatment for PPH reflects primarily the lack of capacity to perform emergency transfusions at a bEmOC facility and increased personnel and facility charges at cEmOC facilities.

Category (bEmOC) Component cost (2006 US$)
Drugs In comparison, Borghi et al. [2003] report the cost for treatment of PPH in Benin and Ghana and Weissman et al. [1999] for Uganda. When these costs are converted to 2006 US$ using the methods described earlier, our estimates from the RHCTM fall within the plausible bounds reported in the literature, which ranged from $46 to $198.

Management/treatment of puerperal sepsis.
At a bEmOC-level facility, the cost is $32.59; for cases needing transfusion, advanced shock management, and/or surgery at cEmOC, the cost is $74.01. In comparison, Bhat et al. [2009] reported a cost of sepsis requiring cEmOC of $66.70 in Gujarat, closely approximating a weighted average of the proportion requiring basic and comprehensive EmOC.

Category (bEmOC) Component cost (2006 US$)
Drugs In addition, Borghi et al. [2003] report the cost for treatment of sepsis in Benin and Ghana and Weissman et al. [1999] for Uganda. When these costs are converted to 2006 US$ using the methods described earlier, our estimates fall within this range.

Treatment of long-term complications such as PID and obstetric fistula
India-specific costs for most long-term sequelae stemming from maternal complications are not available. In their absence we relied on published cost estimates from a variety of developing and developed countries, also used in a previous analysis. [Hu 2007]

Treatment of pelvic inflammatory disease
The cost associated with treating PID is $4.83, assuming all PID is treated on an outpatient basis.

Costs of scaling up and costs in sensitivity analysis
As shown in the figure below, investments for strategies that included stepwise improvements in intrapartum care fall into the following general categories: (1) average normal delivery (differentiated by site) to reflect (a) recruiting and training cadre of SBA, (b) improving recognition of referral need via training of SBA, as well as education for woman and family, and (c) interim care by SBA prior to transport; (2) transfer from delivery site to referral facility (differentiated by origin and destination) to reflect (a) transport cost; (b) vehicle use and fuel; (c) interim care en route separate from routine SBA training; (3) expedient attention at appropriate referral facility (differentiated according to bEmOC or cEmOC services) to reflect (a) new and/or improvements in existing primary facilities (bEmOC) including ensuring 24-hour access; (b) new/and or improved secondary and tertiary facilities (cEmOC); (c) blood bank and transfusion capability, enhanced surgical capacity, intensive care support functions for shock in cEmOC; (d) improved quality of care in bEmOC and cEmOC with adequate supplies and personnel. Deliv ery (birthing centre) Delivery (health c entre with bEmOC) Deliv ery (secondary/tertiary facility-cEmOC) Management of c omplication at bEmOC Management of complic ation cEmOC Transport (home to birthing c entre) T rans port (home to primary facility-bEmOC) Transport (home t o t ertiary facility-cEmOC) T rans port (primary to tertiary fac ility-cEmOC)* Transport (birthing centre to EmO C)

Comparison of scale-up cost implications
There are virtually no empiric data to inform the incremental costs of a massive effort to invest in infrastructure, human resources, and facilities to improve maternal health. Since our incremental "upgrade" costs were at best, rough approximations, we felt it was prudent describe these costs in the context of projected estimates made by others. Borghi 2006;Johns 2007). We also leveraged published studies that estimated the global resource requirements for scaling up interventions to reduce maternal mortality to assess the face validity of our assumptions. (Borghi 2006;Johns 2007) We outline the findings from selected comparisons below.
The Gujarat state developed a public-private partnership called the Chiranjeevi Scheme, which focuses on institutional delivery and emergency obstetric care for the poor. Although one of the most important outcomes was financial protection from catastrophic costs associated with a complication, a range of cost proxies (for resources required) were provided. ] The estimate of costs associated with complications in delivery was $209.33 (among the participants who had a prior complicated delivery). The average cost of complicated deliveries in our scaleup scenarios, which most closely approximate the level of care reported by Bhat et al.
[2009], ranged from $172.58 to $207.33. Johns et al. [2007] found that increasing coverage scale-up targets by 27% (from an average of 73% coverage to 95%) resulted in a 42% increase in total costs. We found that our assumptions of increased costs for later phase improvements (e.g., more than 50% of the population covered), implied that a 33% increase in coverage resulted in an approximate 40% increase in total costs and a 36% increase in discounted lifetime costs. Johns et al. [2007] found that overall, the primary-care level comprises 29% of costs for moderate scale-up and 30% for rapid scale-up, with referral care accounting for 44% of costs in the moderate scale-up scenario and 47% in the rapid scale-up scenario. The remaining 27% in the moderate scale-up scenario and 23% in the rapid scale-up cover programme development and investments in health infrastructure. The relative costs of secondary versus primary costs were 1.57 times higher. We found that our assumptions of increased costs for stepwise improvements implied costs of referral care in secondary and tertiary facilities were 1.56 times the costs of primary-care level delivery and management. Johns et al. [2007] found that overall, the cost of transport was 5% of the total cost of scale-up, and infrastructure comprised 15% of the total cost of scale-up. We found that our assumptions of increased costs for transport from birthing centres to secondary and tertiary facilities ranged from 4.3% to 6% of the total cost of scale-up and transport from home to secondary and tertiary facilities ranged from 4.7% to 12%. Upgrade 1 in urban Uttar Pradesh: 45% facility births; 25% SBA (home births); transport from home (50%), primary-level health centre (60%), bEmOC (70%); recognition of referral need at home (40%), primary-level health centre (60%); availability and quality of EmOC (70%) c Upgrade 3 in urban Uttar Pradesh: 75% facility births; 50% SBA (home births); transport from home (70%), primary-level health centre (80%), bEmOC (90%); recognition of referral need at home (75%), primary-level health centre (90%); availability and quality of EmOC (90%) d Upgrade 1 in rural Uttar Pradesh: 40% facility births; 40% SBA (home births); transport from home (25%), primary-level health centre (60%), bEmOC (75%); recognition of referral need at home (50%), primary-level health centre (75%); availability and quality of EmOC (70%) e Upgrade 3 in rural Uttar Pradesh: 75% facility births; 50% SBA (home births); transport from home (45%), primary-level health centre (80%), bEmOC (90%); recognition of referral need at home (75%), primary-level health centre (90%); availability and quality of EmOC (90%) f Upgrade 1 in Rajasthan: 45% facility births; 25% SBA (home births); transport from home (50%), primarylevel health centre (60%), bEmOC (70%); recognition of referral need at home (40%), primary-level health centre (60%); availability and quality of EmOC (70%) g Upgrade 3 in Rajasthan: 75% facility births; 50% SBA (home births); transport from home (70%), primarylevel health centre (80%), bEmOC (90%); recognition of referral need at home (75%), primary-level health centre (90%); availability and quality of EmOC (90%)

Supplemental Results: Family Planning
Sensitivity analyses that reduce unmet need for spacing versus limiting, without targeting specific age groups, had minimal overall policy impact when varied across plausible ranges. In contrast, focused family planning interventions that increase uptake of contraception in younger women (and therefore weighted towards spacing) had a greater impact than those focused on uptake in older women (and therefore weighted towards limiting). The effect of focused family planning interventions that shift contraceptive choices with higher failure rates to more effective options had minimal effect on the overarching policy choices in large part because of the small proportion of women affected. We also conducted an analysis exploring family options for reducing the unmet need for limiting, spacing and shifting to more effective contraception for women who experienced pregnancy (i.e., failure) in Uttar Pradesh and Rajasthan.   Status quo: 31.1% facility births (70% in facility with incomplete bEmOC capabilities); 68.9% home births; 11.6% SBA (home births); transport from home (24.4%), primary-level health centre (48.8%), bEmOC (61%); recognition of referral need at home (20%), primary-level health centre (40%); availability and quality of EmOC (42.5%).
c Antenatal care only assumes no increased services as part of antenatal care, and no linkages with enhanced interventions that might be related.
d Antenatal care with effective treatment for anemia has a greater effect, with up to 5.5% mortality reduction independent of any other interventions.
e Antenatal care that then increases the likelihood of a facility-based delivery (30% in bEmOC capable facility and 70% in health centre) more effective and cost-effective, at 50% to 75% the per capita GDP.

Health Centre and Birthing Centres, bEmOC and cEmOC
While the main analysis in the paper presents a number of scenarios that incrementally shift women from home deliveries to facility-based deliveries, here we conduct a sensitivity analysis varying assumptions with respect to the distribution of facilities. The example provided below assumes that 60% of deliveries are facility-based outside the home. In Table 4A we varied assumptions (depicted in yellow shading) about the proportion in birthing centres as compared to EmOC facilities while holding the distribution within EmOC facilities constant (90% bEmOC and 10% cEmOC with referral to cEmOC when necessary. We assumed 70% effective referral from the birthing centre and 80% effective referral from bEmOC to cEmOC based on the availability of transport. Results are shown (pink shading) below. Largely due to imperfect availability of prompt transport and accurate referral from the health centre, mortality is reduced most when deliveries occur in EmOC facilities although by only a small margin. Accordingly, given the cost advantage of birthing centres, the incremental cost-effectiveness ratios for Scenario 1 are low (i.e., very attractive), and in fact less than 25% of the per capita GDP. Cost-effectiveness ratios increase from $240 to $810 if one restricts routine deliveries to EmOC facilities only. Of note, provided the EmOC facilities are mainly bEmOC, the ratio is still less than the per capita GDP and thus this would be considered cost-effective a. HC = health centre, BC = birthing centre. bEmOC = basic emergency obstetric care, cEmOC = comprehensive emergency obstetric care. HC and BC used interchangeably and assume SBA, clean delivery, expectant management, but lack all 6 signal functions. In this particular analysis we also assume SBA-administered misoprostol in birthing centres/health centres. b. Scenario 1: Assumes facility-based delivery 70% in HC/BC and 30% in EmOC facilities. c. Scenario 4: Assumes facility-based delivery all occurs in EmOC facilities. d. Incremental cost-effectiveness ratio calculated as the additional costs divided by the additional health effects.

62
In Table 4B we varied assumptions about the proportion of routine deliveries in bEmOC versus cEmOC. In this analysis we assumed that transport/succesful referral was available for 95% of women delivering in bEmOC who required it. We assumed 70% effective referral from the birthing centre, as we did above. Results for shifting the base case distribution (90% bEmOC and 10% cEmOC, as shown in blue) to a distribution where the majority of routine deliveries occur in cEmOC, are shown in green. As expected, this change results in cost-effectiveness ratios that are clearly well beyond traditionally acceptable thresholds for cost-effectiveness, and they became less attractive as one shifts a greater number of births to EmOC (e.g., right side of the table, scenario 4). In fact, this is one of the least efficient strategies that we identified. 10% bEmOC 90% cEmOC $8,300 b $13,800 b $20,700 b $27,700 b a. HC = health centre, BC = birthing centre. bEmOC = basic emergency obstetric care, cEmOC = comprehensive emergency obstetric care. HC and BC used interchangeably and assume SBA, clean delivery, expectant management, but lack all 6 signal functions. In this particular analysis we also assume SBA-administered misoprostol in birthing centres/health centres. b. Incremental cost-effectiveness ratio calculated as the additional costs divided by the additional health effects, in comparison to the base case 90%bEmOC and 10%cEmOC.

Supplemental Results Table 5. Sensitivity Analysis: Comparison of Community-Based Intervention Analysis (Pagel)
We compared our model projected benefits of a community based intervention where antibiotics for sepsis, and misoprostol for PPH were added to births at health centers, and at home when skilled attendants were present to those reported in Pagel et al (2009). We approximated the coverage levels of antibiotics for sepsis and misoprostol distribution (and approximate percentage of facility births) by applying the intervention to our model of rural India. Strategies were similar but not identical as our model includes all causes of maternal mortality, multiple facility-levels, mild, moderate and severe complications, and explicit consideration of the three delays.

64
Coverage range (Pagel et al. 2009 would only receive oxytocin if she gave birth in a facility. In our model there are multiple levels of facilities, primary, secondary and tertiary and oxytocin is available in secondary and tertiary facilities. In the base case, birthing centres are intended to be interim primary-level facilities staffed by SBA but not with full bEmOC capabilities. Our analysis of misoprostol assumes use by SBA for both home deliveries and for birthing centre deliveries. b In Pagel et al, if a woman develops an infection after delivery she could obtain antibiotics either from a facility or privately. In our model a woman would only receive antibiotics if she delivers in or is successfully referred to a facility capable of providing intrapartum care. In both Pagel et al and our model, the effectiveness of antibiotics for reducing sepsis related deaths in the community setting was 85%. c Package 1 strengthens facilities, ensuring that more facilities have available oxytocin and antibiotics, package 2 combines package 1 with distribution of misoprostol to women attending outreach antenatal care appointments (provided drug was available) and distribution of antibiotics by CHW to women with signs of postnatal infection, and package 3 enhances package two with additional misoprostol and antibiotics via female volunteers in villages. d In Pagel et al, the probability that a woman would take misoprostol if it was given outside of a facility was 85%, and the probability facility would have uterotonic drugs in packages 1-3 was 95%. We assume that facility based availability of uterotonics is a function of quality and availability of services which were varied across upgrades from 42% to 95%. The relative risk reduction for severe PPH with misoprostol was 0.61 in Pagel et al, and was varied from 0.50 -0.61 in our model. Pagel et al (2009) reported annual prevented deaths from PPH and sepsis in both percentage reductions and absolute cases for Malawi and Sub-Saharan Africa. Our model projects the number of deaths that would be averted over the lifetime of a birth cohort of 100,000 girls, as well as the reductions achieved from these interventions. Further, our model includes competing risk from all other causes of maternal mortality, and includes both severe and moderate or less PPH. Correcting for this, we show below the results when expressed in the format projected from Pagel et al (2009). While the absolute numbers are not directly comparable given population differences, the magnitude of mortality risk reduction for these specific community-based interventions is comparable. Interestingly, the results for the lowest three quintiles reported in Pagel et al (2009)