The authors have declared that no competing interests exist.
Conceived and designed the experiments: AR TTB SS PB. Performed the experiments: AR TTB. Analyzed the data: AR NG. Contributed reagents/materials/analysis tools: ES PB. Wrote the paper: AR TTB SS PB.
Despite increasing access to antiretroviral treatment (ART) in low-income countries, HIV-related mortality is high, especially in the first months following ART initiation. We aimed to evaluate the impact of TB coinfection on early mortality and to assess gender-specific predictors of mortality in a cohort of Ethiopian adults subjected to intensified casefinding for active TB before starting ART.
Prospectively recruited ART-eligible adults (n = 812, 58.6% female) at five Ethiopian health centers were followed for 6 months. At inclusion sputum culture, Xpert MTB/RIF, and smear microscopy were performed (158/812 [19.5%] had TB). Primary outcome was all-cause mortality. We used multivariate Cox models to identify predictors of mortality.
In total, 37/812 (4.6%) participants died, 12 (32.4%) of whom had TB. Karnofsky performance score (KPS) and mid-upper arm circumference (MUAC) were associated with mortality in the whole population. However, the associations were different in men and women. In men, only MUAC remained associated with mortality (adjusted hazard ratio [aHR] 0.71 [95% CI 0.57–0.88]). In women, KPS <80% was associated with mortality (aHR 10.95 [95% CI 2.33–51.49]), as well as presence of cough (aHR 3.98 [95% CI 1.10–14.36]). Cough was also associated with mortality for TB cases (aHR 8.30 [95% CI 1.06–65.14]), but not for non-TB cases.
In HIV-positive Ethiopian adults managed at health centers, mortality was associated with reduced performance score and malnutrition, with different distribution with regard to gender and TB coinfection. These robust variables could be used at clinic registration to identify persons at increased risk of early mortality.
Although survival in people living with HIV (PLWH) has increased during the last decade as a result of expanding access to antiretroviral treatment (ART), HIV-related mortality remains high, especially in sub-Saharan Africa where 800,000 deaths were caused by HIV/AIDS in 2014 [
Despite these revised criteria for ART, large proportions of PLWH still present with advanced immunosuppression at HIV diagnosis. According to a recent meta-analysis estimating temporal trends in disease status at presentation to care and ART initiation, mean CD4 cell counts did not increase significantly from 2002 to 2012 in sub-Saharan Africa [
Besides CD4 cell count at ART initiation, other factors may be involved in HIV-related mortality. For example, low body-mass index (BMI), anaemia, and increasing age have been associated with elevated risk of death in PLWH starting ART [
Importantly, the underlying causes of death in PLWH show regional variations, and are incompletely understood. In sub-Saharan Africa, most fatalities in PLWH are considered to be caused by tuberculosis (TB) [
We have previously presented data on TB coinfection and ART outcomes in a cohort of treatment-naïve HIV-positive adults who met criteria for ART recruited at Ethiopian health centers [
The cohort used for this study was recruited to determine the prevalence of active TB in ART eligible patients at Ethiopian health centers, to evaluate different diagnostic methods for TB and to determine the outcome of ART and TB treatment. The cohort and study setting have previously been described in detail [
At inclusion, trained non-physician clinicians collected detailed clinical data, including findings from physical examination, following a structured questionnaire. All participants (irrespective of symptoms) were requested to provide sputum samples for TB diagnostics by liquid culture, GeneXpert MTB/RIF, and smear microscopy. In case of peripheral lymphadenopathy, TB diagnostics were performed on fine-needle aspirates. Patients who did not submit sputum samples were excluded from the cohort since they could not be categorized for TB. Blood samples were obtained for complete blood and CD4 cell counts. Results from blood tests and TB investigations were available to the responsible clinicians. Detailed description of the TB diagnostic procedure has been published previously [
Follow-up visits were scheduled at months 1, 2, 3, and 6 after study inclusion. In case of symptoms or signs suggestive of TB during follow-up study clinicians were encouraged to repeat TB diagnostics according to the baseline protocol. Health extension workers traced patients with missed appointments by telephone and home visits. The health center clinicians were responsible for starting ART and treating opportunistic infections (including TB) in accordance with national guidelines [
The primary outcome for this study was all-cause mortality within 6 months of study inclusion. Analysis of predictors for this outcome was based on characteristics collected at study inclusion. This analysis was first performed for the whole population, and then separately with regard to gender and TB status, respectively. Time-at-risk was defined as days from inclusion until time of death or 6 months after inclusion. Patients more than 3 months late for a scheduled visit were defined as lost to follow-up (LTFU) and their follow-up time was right censored at the last study visit. For patients who transferred their care to other facilities or who declined further follow-up similar right censoring was used.
Patient characteristics were summarized using frequencies and percentages for categorical variables and median and interquartile range (IQR) for continuous variables. For all continuous variables, plots of beta estimates for the primary outcome for each quintile of the variable were assessed to identify each variable’s functional form in order to define appropriate cut-off values. Mid-upper arm circumference (MUAC) and BMI were kept as continuous variables. CD4 cell count was categorized by quartiles, and haemoglobin according to the WHO anaemia cut-offs, merging no anaemia (haemoglobin ≥13.0 and ≥12.0 g/dL for men and women, respectively) and mild anaemia (haemoglobin 11.0–12.9 g/dL and 11.0–11.9 g/dL for men and women, respectively). Karnofsky performance score (KPS) was dichotomized at <80% differentiating those unable to carry out normal activities or do active work. Patient characteristics for male and female participants, as well as for TB cases and non-TB cases, were compared using χ2-test for categorical, and Mann-Whitney’s U-test for continuous variables.
Kaplan-Meier survival analyses were used to investigate overall survival from inclusion, using the log rank test to compare male and female participants as well as TB cases and non-TB cases, respectively. Cox proportional hazards models were used to investigate baseline variables’ correlation with mortality. In order to further assess associations with mortality, four separate models were fitted for males, females, TB cases, and non-TB cases, respectively. All variables considered for the Cox models were assessed for the proportional hazards assumption using Kaplan-Meier and log minus log plots, and by analysing each variables’ interaction with time. Only variables fulfilling the assumption were included in the final models.
Variables from each univariate model with a p value <0.3 were entered into multivariate models, whereby the least significant variables were removed using a backward stepwise procedure until only variables with p<0.05 remained in the model. At each step, the beta estimates of variables remaining in the model were analysed for possible interaction with the variable removed. A change in beta estimate of >20% was considered to indicate a possible interaction, which then was further evaluated. The multivariate models were adjusted for age and CD4 cell count, as well as ART status, included as a time-varying 0/1 variable. Time-to-ART was defined as days from study inclusion until start of ART.
To account for missing data, secondary analyses were performed in which the multivariate models were refitted excluding variables with <95% available data from the start of the stepwise procedure. The final model generated was then compared with the original model to assess the possible bias derived from cases with missing data.
For all analyses, a p value <0.05 was considered to indicate statistical significance. All analyses were performed using SPSS, version 21 (IBM Corp, Armonk, NY).
Ethical approval was obtained from the National Research Ethics Committee at the Ministry of Science and Technology of Ethiopia and the Regional Ethical Review Board of Lund University, Sweden. All study participants provided written informed consent.
The health center clinicians were responsible for all aspects of HIV and TB care in line with national guidelines. All laboratory results were communicated back to the responsible clinician, including all TB results. No treatments were withheld from the patients due to this study.
A total of 886 PLWH were screened for eligibility; 74 were excluded and 812 participants (58.6% female) were included in the study cohort (
All n = 812 | Male n = 336 | Female n = 476 | TB cases n = 158 | Non-TB cases n = 654 | |||
---|---|---|---|---|---|---|---|
32 (28–40) | 35 (30–42) | 30 (26–36) | 34 (28–40) | 32 (28–39) | 0.18 | ||
476 (59) | – | – | 75 (47) | 401 (61) | |||
18.9 (17.3–21.0) | 18.8 (17.2–20.3) | 19.1 (17.4–21.6) | 17.7 (16.3–19.6) | 19.2 (17.6–21.3) | |||
23 (21–24) | 23 (21–24) | 22 (20–24) | 0.51 | 21 (19–23) | 23 (21–25) | ||
208 (116–320) | 191 (104–302) | 219 (132–331) | 170 (91–273) | 220 (127–328) | |||
229 (28) | 85 (26) | 144 (30) | 33 (21) | 196 (30) | |||
197 (24) | 76 (23) | 121 (26) | 27 (17) | 170 (26) | |||
223 (28) | 93 (28) | 130 (27) | 53 (34) | 170 (26) | |||
158 (20) | 78 (24) | 80 (17) | 45 (28) | 113 (17) | |||
12 (8–17) | 11 (7–16) | 13 (9–17) | 10 (9–12) | 12 (8–17) | 0.41 | ||
158 (19) | 83 (25) | 75 (16) | – | – | |||
0.11 | |||||||
467 (61) | 206 (66) | 261 (58) | 56 (38) | 411 (67) | |||
268 (35) | 97 (31) | 171 (38) | 79 (53) | 189 (31) | |||
27 (4) | 11 (4) | 16 (4) | 14 (9) | 13 (2) | |||
557 (69) | 218 (65) | 339 (72) | 0.05 | 93 (59) | 464 (71) | ||
442 (55) | 205 (61) | 237 (50) | 107 (68) | 335 (51) | |||
514 (64) | 238 (71) | 276 (58) | 130 (82) | 384 (59) | |||
401 (50) | 168 (50) | 233 (49) | 0.71 | 101 (64) | 300 (46) | ||
553 (69) | 233 (71) | 320 (69) | 0.56 | 131 (85) | 422 (66) | ||
389 (48) | 165 (49) | 224 (47) | 0.58 | 100 (64) | 289 (44) | ||
326 (40) | 149 (44) | 177 (37) | 98 (62) | 228 (35) | |||
396 (49) | 174 (52) | 222 (47) | 0.16 | 113 (72) | 283 (43) | ||
126 (16) | 58 (18) | 68 (14) | 0.22 | 41 (26) | 85 (13) | ||
275 (34) | 120 (36) | 155 (33) | 0.35 | 89 (56) | 186 (28) | ||
141 (17) | 54 (16) | 87 (18) | 13 (8) | 128 (20) | |||
246 (30) | 84 (25) | 162 (34) | 31 (20) | 215 (33) | |||
322 (40) | 146 (44) | 176 (37) | 76 (48) | 246 (38) | |||
100 (12) | 50 (15) | 50 (11) | 28 (24) | 62 (10) |
Data presented as n (%), or median (interquartile range). Data available for >97% for all variables, except haemoglobin with 762/812 (93.8%) available data.
Abbreviations: BMI, body-mass index; MUAC, mid-upper arm circumference; TB, tuberculosis.
1
2
At 6 months after inclusion 679 participants (83.6%) remained in care, 37 (4.6%) were confirmed dead, 42 (5.2%) LTFU, 35 (4.3%) had registered transfer of care, and 20 (2.5%) declined further follow-up. ART was started in 564 (69.5%) subjects a median of 27 days (IQR 12–57) after inclusion. The median time to death for the 37 deceased participants was 54 days (IQR 30–87); 18 (48.6%) died before and 19 (51.4%) died after starting ART. In the latter group, death occurred a median of 51 days (IQR 33–72) after starting ART. In total, 12 (32.4%) of the deceased participants had been diagnosed with TB; 8 (66.7%) of these had not started ART and 4 (33.3%) died after having started ART (
Mortality | |||
---|---|---|---|
Total | Before ART start | After ART start | |
37/812 (4.6) | 18 (48.6) | 19 (51.4) | |
By gender: | |||
20/336 (6.0) | 9 (45.0) | 11 (55.0) | |
17/476 (3.6) | 9 (52.9) | 8 (47.1) | |
By TB status: | |||
12/158 (7.6) | 8 (66.7) | 4 (33.3) | |
25/654 (3.8) | 10 (40.0) | 15 (60.0) |
Presented as n/N (%) and n (%). Abbreviations: ART, antiretroviral treatment.
16 (50.0%) of TB cases died before starting TB treatment.
No significant difference was found when comparing survival by gender (p = 0.10) using Kaplan-Meier plots. Non-TB cases had significantly better survival than TB cases (p = 0.03,
Complete univariate hazard ratios for all tested variables are presented in
In the multivariate analysis for the entire cohort, KPS<80% and MUAC were independently associated with mortality (
Subgroup and variables | aHR (95% CI) |
---|---|
Karnofsky score <80% | 4.30 (1.84–10.08) |
MUAC—per cm increase | 0.82 (0.71–0.94) |
MUAC—per cm increase | 0.71 (0.57–0.88) |
CD4 cell count—cells/μL | |
>300 | 1.00 |
201–300 | 1.82 (0.25–13.23) |
100–200 | 1.95 (0.34–11.18) |
<100 | 6.80 (1.37–33.78) |
Karnofsky score <80% | 10.95 (2.33–51.49) |
Reported cough | 3.98 (1.10–14.36) |
Reported cough | 8.30 (1.06–65.14) |
Gender, male vs. female | 2.53 (1.05–6.12) |
Karnofsky score <80% | 4.89 (1.81–13.19) |
MUAC—per cm increase | 0.76 (0.63–0.91) |
Abbreviations: aHR, adjusted hazard ratio; CI, confidence interval; MUAC, mid-upper arm circumference.
Adjusted for age as a continuous variable, CD4 cell counts as a categorical variable and antiretroviral treatment as a time-varying variable.
Among men MUAC had a stronger negative association with mortality, aHR 0.71 (95% CI 0.57–0.88); however, KPS<80 was not associated with mortality in men. Furthermore, male participants with CD4 cell counts <100 cells/μL had an aHR of 6.80 (95% CI 1.37–33.78) compared with CD4 cell counts >300 cells/μL.
For women KPS<80% had the overall strongest correlation with mortality with aHR 10.95 (95% CI 2.33–51.49), but the association between MUAC and mortality did not reach statistical significance. Reported cough at study inclusion was independently associated with an increased risk of early mortality for female participants, aHR 3.98 (1.10–14.36).
In the multivariate model for TB cases, reported cough was the only variable that remained in the final model, aHR 8.30 (1.06–65.14).
Among participants not diagnosed with TB male gender was associated with mortality (aHR 2.53 [95% CI 1.05–6.12]), as well as KPS<80% aHR 4.89 (1.81–13.19) and MUAC aHR 0.76 (95% CI 0.63–0.91) per centimeter increase (
Haemoglobin was missing for 50 (6.2%) participants. Therefore, all multivariate models were refitted excluding anaemia from the start. The final models did not change.
In this cohort of HIV-positive Ethiopian adults, mortality during the first 6 months after inclusion was associated with reduced performance score and malnutrition (assessed by MUAC measurement). As part of the cohort study protocol, all participants had been investigated for active TB at inclusion. Although this resulted in detection of TB in 19.5% of subjects, persons with TB were at increased risk of death.
In contrast to other studies [
In this study, time-at-risk started at the time of study inclusion and we did not restrict the analysis of mortality to those who actually started ART. This design allowed us to estimate the proportion and risk factors for death also in ART-eligible persons who died before starting treatment. The median time between inclusion and death for deceased subjects was short (48 days), and they had advanced disease at presentation. In particular, a greater proportion had TB [
In the current WHO guidelines all PLWH should start ART [
In this study we aimed to include potential markers of mortality risk that would be possible to use at health center level in low-income countries. Based on previous findings from our own studies and on the scientific literature, we hypothesized that poor performance status and malnutrition would be associated with mortality. Karnofsky performance score was chosen as we wanted a general, standardized, and established measure of the performance status of participants that could be easily recorded by health care workers. Malnutrition can be measured using BMI and/or MUAC. Both our research group and others have shown independent association between MUAC and mortality in patients with TB [
Unfortunately, causes of death were not determined for our participants. This reflects the usual situation in low-income countries, and highlights the importance of detailed post-mortem investigation studies. We could ascertain, primarily by telephone calls to the family of the deceased, that the deaths in our study were due to illness and not attributable to accidents or other unnatural causes. It is possible that use of verbal autopsies could have elucidated causes of death with some greater accuracy [
In our cohort, all participants had undergone active TB casefinding at inclusion, leading to diagnosis of TB in 158/812 persons, which in most cases (91.8%) had not been identified previously. Among participants not diagnosed with TB in our cohort, mortality was associated with male gender, low KPS and MUAC. Interestingly, these latter variables have also been found to be independently associated with prevalent TB in this cohort [
This study was based on a large material of patients with structured and detailed categorization for active TB at inclusion. Furthermore, participants were recruited at Ethiopian public health centers, representing a typical health care setting for where most PLWH receive care in low-income countries.
Our study has certain limitations. For a considerable proportion of patients the outcome at 6 months was unknown (LTFU 42/812; 5.2%). LTFU is common in ART programs in sub-Saharan Africa, and studies on LTFU have identified different reasons for this phenomenon [
In conclusion, we found that two simple clinical measurements have strong independent association with 6-month mortality in Ethiopian PLWH eligible to start ART: MUAC for men, and KPS for women. These variables could be considered for use in routine care to identify subjects at high risk of early mortality. Such persons may particularly benefit from fast-track ART initiation, as well as intensified investigation for TB.
(SAV)
(DOCX)
We extend our gratitude to the patients who participated in the study as well as to the staff members at the health centers and the Adama Regional laboratory for their work with this study. We also acknowledge our data management team: Gadissa Merga and Surafel Girma, who contributed greatly to this study. We are also grateful for the excellent collaboration with the Oromia Regional Health Bureau. This study is registered with clinicaltrial.gov, number NCT01433796.