Return on investment of self-measured blood pressure is associated with its use in preventing false diagnoses, not monitoring hypertension

Previous research indicates that patient self-measured blood pressure (SMBP) is a cost-effective strategy for improving hypertension (HTN) diagnosis and control. However, it is unknown which specific uses of SMBP produce the most value. Our goal is to estimate, from an insurance perspective, the return-on-investment (ROI) and net present value associated with coverage of SMBP devices when used (a) only to diagnose HTN, (b) only to select and titrate medication, (c) only to monitor HTN treatment, or (d) as a bundle with all three uses combined. We employed national sample of claims data, Framingham risk predictions, and published sensitivity-specificity values of SMBP and clinic blood-pressure measurement to extend a previously-developed local decision-analytic simulation model. We then used the extended model to determine which uses of SMBP produce the most economic value when scaled to the U.S. adult population. We found that coverage of SMBP devices yielded positive ROIs for insurers in the short-run and at lifetime horizon when the three uses of SMBP were considered together. When each use was evaluated separately, positive returns were seen when SMBP was used for diagnosis or for medication selection and titration. However, returns were negative when SMBP was used exclusively to monitor HTN treatment. When scaled to the U.S. population, adoption of SMBP would prevent nearly 16.5 million false positive HTN diagnoses, thereby improving quality of care while saving insurance plans $254 per member. A strong economic case exists for insurers to cover the cost of SMBP devices, but it matters how devices are used.


Model
The model consists of two primary parts: Part 1, characterized by a decision tree model, describes the diagnosis, treatment selection and management decision process. Part 2, characterized by a Markov model, describes the transition to different cardiovascular disease (CVD) events (stroke, myocardial infarction or other CVD events). Seven age groups are modeled: 25 to 34, 35 to 44, 45 to 54, 55 to 64, 65 to 74, 75 to 84, and 85 or more years old. Individuals within each age group are assumed to have the same health characteristics and to be insured by a commercial private insurance. All insured individuals go through part 1 of the model in year 1, and then move and remain in part 2 of the model until they die or leave the health insurance plan. The model assumes annual cycles. All costs and returns are in 2016 US dollars and were discounted at a 3% annual discount rate. All simulations were performed in Microsoft Excel.  Table A).

Decision tree model
Pathway 1: Individuals previously screened and given a diagnosis of hypertension (DxHT). Based on known CBPM sensitivity and specificity (1), we assumed that a fraction of these individuals, including those with white-coat hypertension (WCH) who exhibit high blood pressure in the clinic but not outside the clinic (2), would be diagnosed incorrectly and we classified them as false positives (FP). The remaining fraction was assumed to be correctly diagnosed and were classified as true positives (TP). The model mirrors usual practice by assuming that these individuals are not re-evaluated with SMBP, since this would necessitate taking them off medications. The model also assumes that some of those classified as hypertensive (whether or not the diagnosis was correct) would adhere to recommended treatment (Tx) and some would not (~Tx) (3,4).
Pathway 2: Individuals previously screened and determined to be normotensive (DxNT). A fraction of these, including individuals with masked hypertension whose BP is normal when measured in the clinic but high outside the clinic (5,6), were assumed to be incorrectly diagnosed and were classified as false negatives (FN). The remaining fraction of normotensive individuals were assumed to be correctly diagnosed and were classified as true negatives (TN). We assumed these individuals would be re-evaluated with SMBP and those confirmed to be normotensive would not receive treatment.
Pathway 3: Individuals with no screening or diagnostic history and whose diagnosis is unknown (~Dx). We assumed these individuals would be evaluated with SMBP.
(b) The true prevalence of hypertension is unobserved and no intervention is perfectly accurate in detecting hypertension. Among those diagnosed as hypertensive, we assumed, based on published sensitivity and specificity estimates for SMBP or usual care via CBPM (1), that a fraction was incorrectly diagnosed (false positives, FP) and the remainder were correctly diagnosed (true positives, TP). Similarly, among those diagnosed as normotensive, we assumed a fraction was incorrectly diagnosed (false negatives, FN), and the remainder were accurately diagnosed (true negatives, TN). See Table A for diagnosis accuracy rates.
(c) The model assumes that those who are normotensive (DxNT) or whose diagnosis is unknown (~Dx) would be diagnosed based on either SMBP or CBPM and be classified as hypertensive (DxHT) or normotensive (DxNT). However, those already diagnosed with hypertension (DxHT) based on CBPM would not have their diagnosis changed, since this would involve taking them off medication; therefore, these individuals would not be re-classified based on the SMBP intervention.
(d) After diagnosis, all individuals classified as hypertensives (DxHT) are offered antihypertensive medication regardless of their true hypertension status. A fraction of those (see Table A) adhere to the selected medication treatment (Tx), while the rest do not adhere (~Tx).
(e) Normotensive individuals (DxNT) do not receive anti-hypertensive medication (~Tx), regardless of their true hypertension status.
(f) Regardless of their diagnosis (DxHT or DxNT) and adherence to medication (Tx or ~Tx), all individuals who are truly hypertensives (TP and FN) are at risk of developing a hypertensionrelated CVD event, and therefore, at the end of the first year, they will move to the second part of the model (Markov).  In that year and all subsequent years, a fraction of individuals will suffer a CVD event, a fraction will die or leave the insurance plan, and therefore, exit the model permanently, and a fraction will continue to the following year free of hypertension-related CVD risk.
(g) The fraction of individuals who transit to a CVD event state will do so with a probability based on their Framingham-based CVD (stroke, myocardial infarction or other CVD) incidence rate.
(h) The CVD state represents three CVD events: stroke, myocardial infarction or other CVD event. All individuals will remain in this state for one year until they move to a post-CVD state or they exit the model due to a death event or leaving the insurance plan.
(i) Transit from a CVD to a post-CVD state occurs in one year for all surviving individuals to reflect the changes in their risk factors and costs after a CVD event.
(j) The post-CVD state represents the long-term recuperation period for stroke, myocardial infarction or other CVD events. Individuals will stay in this state for up to 5 years. However, during that 5-year period, they can suffer a subsequent CVD event, in which case they recycle, or they can exit the model.
(k) Individuals in the post-CVD state remain there for up to five years to reflect the time it takes to converge to baseline costs and health status at the ~CVD state.
(l) A fraction of individuals will suffer other CVD event within the first five years after their first CVD event. Subsequent events occur with an age-dependent incidence probability of a stroke, myocardial infarction or other CVD event.
(m) After five years in the post-CVD state, individuals who do not suffer a subsequent CVD event return to a baseline state of CVD risk.
(n) At any point in time, any individual can exit the model by i) dying from a non-CVD related event, ii) dying after a CVD event, or iii) leaving the health insurance plan. Both, death and leaving the plan are absorbing states, meaning that individuals in those states leave the model permanently.

Interventions and use cases
We model and compare two interventions: the SMBP intervention and usual care. For the SMBP intervention we assume patients receive a clinically validated SMBP device and educational support for its appropriate use. We also assume that physicians follow recommended procedures for SMBP-based diagnosis and treatment selection (7). For BP monitoring (management of hypertension), we assume that all patients with hypertension keep the SMBP device. We assume further that the device is replaced every 5 years, and educational support regarding its use is provided annually. For usual care, we assume that hypertension is diagnosed using CBPM, with no additional BP measurements outside the clinic.
We distinguish three different use cases for SMBP: i) Diagnosis: Use of SMBP to establish the diagnosis of hypertension. This use applies only to individuals seen by a doctor within the previous two years and found to be normotensive, and to individuals whose diagnosis is unknown. Individuals whose diagnosis was previously established using CBPM were excluded from this case, since we assumed that these individuals would not be taken off medication and re-evaluated with SMBP ii) Treatment selection and titration: Use of SMBP to select an appropriate anti-hypertensive treatment and titrate the dosage. We assumed that physicians requiring four or more visits to select and titrate treatment using CBPM, would need one fewer visit if they used SMBP. This assumption was based on an episode-of-care analysis of claims data (see below for details) where we found that 90% of the treatment-selection-and-titration episodes required 4 or fewer office visits when CBPM was used. Our model assumes that SMBP would reduce excessive visits among the remaining 10% that require more than 4 visits. We performed a sensitivity analysis for this assumption.
iii) BP Monitoring: Use of SMBP for ongoing management of hypertension for BP control following treatment selection. In this case, the diagnosis would have been established previously, with either CMBP or SMBP, the treatment already would have been selected, and patients would now be receiving ongoing monitoring to control their BP.
For the SMBP intervention, we include all three use cases. For usual care, we do not include an out-of-clinic intervention for management of hypertension.
Device accuracies (sensitivity and specificity) and the effects of SMBP on hypertension management were obtained from the literature. We assumed that SMBP is able to improve medication adherence and reduce systolic BP (8,9).

Episode of care analysis
We use an empirical episode-of-care analysis to characterize the diagnosis and antihypertensive treatment selection use cases. Data for these analysis come from the Truven Health MarketScan® Lab Database (10) on commercial insurance claims, 2009 to 2015. Because CBPM is still extensively used in the U.S., we assume that the Truven database represents usual care (diagnosis and monitoring by CBPM). We first identify patients newly diagnosed with essential hypertension based on diagnosis codes, procedure codes, and place of service. An episode of hypertensive care is then identified by considering only patients who had at least 6 months continuous enrollment before the initial hypertension diagnosis date, and at least 18 months continuous enrollment after that date. Individuals are excluded if they had an antihypertensive medication fill during the 6 months prior to the hypertension diagnosis date, since individuals who were already taking antihypertensive medications were likely diagnosed at an earlier time. The remaining patients are included in the population of newly diagnosed hypertension. Antihypertension medication fills are defined as Diuretics, Angiotensin Converting Enzyme (ACE) Inhibitors, Angiotensin Receptor Blockers (ARBs), Beta Blockers, and Calcium Channel Blockers (CCBs).
To identify the end date of an episode of care, we employ a concept of treatment equilibrium. A newly diagnosed patient is assumed to begin an episode of care in a state of disequilibrium (uncontrolled BP). To reach a state of treatment equilibrium (controlled hypertension), the physician must find a medication type and dosage combination that controls the patient's BP without causing adverse drug side effects. The episode of care begins with a new hypertension diagnosis and ends at the point in time following the diagnosis date when the patient is on a stable treatment regimen, i.e., continuously prescribed medication(s). We define a stable regimen, or treatment equilibrium, as four consecutive fills of the same antihypertensive medication and dose, with the fourth prescription fill coming 90 days following the initial hypertension diagnosis date. To determine equilibrium using claims data, we evaluate each medication fill in terms of standardized molecule and dosage per day, based on medication dosage, days' supply, and quantity dispensed. Applying this definition of treatment equilibrium, we determined how many days and how many visits each newly diagnosed patient took to reach BP control, the point of equilibrium, under a usual-care scenario. Not every patient will reach treatment equilibrium. (c) True prevalence and sensitivity and specificity for each technology were also used to determine diagnosis of hypertension.

Data Sources and parameters
(d) We used NHANES 2013-14 to estimate the fraction of individuals with hypertension that are adherent to their medication.

Part 2: Markov model
Following the annotations in Fig B., the following parameters were assumed in the model.
(g) We used the 10-year Framingham risk equation to evaluate the average patient from the simulated cohort. Annual probabilities of CVD events are obtained from the current age of the cohort until age 100. All surviving individuals are assumed to die at age 100. One-year probabilities of CVD events are obtained separately for males and females for each CVD event (MI, stroke and other CVD) based on the 2008-2010 average of CVD distribution by age. Probabilities are then combined based on female-male distribution. Age-dependent probabilities are presented in Table C. For illustration, one-year probabilities of CVD events for the age group 45 to 54 years old are represented in Fig C.   Fig C. One-year probabilities of CVD events for age group 45-54 (SMBP, under treatment) (i) Transition from CVD to post-CVD state occurs after one year for all surviving individuals.
(k) All surviving individuals remain in the post-CVD state for up to five years unless they have a subsequent CVD event, die, or leave the insurance plan.
(l) One-year probabilities of subsequent CVD events were obtained from 4-year follow-up Framingham data on subsequent CVD events (13). One-year probabilities of CVD events were obtained separately for males and females for each CVD event (MI, stroke and other CVD) based on the 2008-2010 average of CVD distribution by age. Probabilities were then combined based on female-male distribution. Age-dependent probabilities are presented in Table C. Fig D shows the 1-year probabilities of subsequent CVD events for the age group 45 to 54 years old.

Fig D. One-year probabilities of subsequent CVD events for age group 45-54 (SMBP)
(m) Return to baseline risk is assumed after 5 years of cardiovascular disease (calculated).

Treatment and CVD costs
Following the annotations in Figs A and B, the following costs were assumed in the model.
(c) For the SMBP intervention, we assume an initial investment in the device, priced at a retail cost of $50, and including additional educational support to the patient paid by insurance (see Table C). When SMBP is used to diagnose and select treatment (Use Cases #1 and #2), a 30minute nurse-provided training session valued at $34.70/hour (18) is included. When SMBP is used for ongoing BP monitoring (Use Case #3), we included one-hour of nurse training annually for educational support to the patient. We also assumed that all patients with hypertension receive an SMBP device for ongoing BP monitoring that is replaced every 5 years. In all use cases, we assumed patients kept the SMBP device and that copayments were not required.
(d) As described in section 2 of this Appendix, we used an episode-of-care approach to characterize the interval encompassing diagnosis and anti-hypertensive treatment selection and dosage titration. We identified all costs associated with a hypertension-related episode of care, including laboratory tests, doctor's visits, and hospital and emergency department visits incurred during the identified episode. Since CBPM, but not SMBP, is still used extensively in the United States to diagnose hypertension, we assumed that cost estimates based on Truven database statistics were representative of usual care (CBPM). For SMBP, potential reductions in utilization were estimated as follows. We assumed that those having 4 or more visits to reach a treatment selection under usual care, will have one fewer visit under SMBP. Using the Truven database, we estimated the average cost of the last visit saved for those with 4 or more visits at $650.6. This estimate translates to an average cost per person of SMBP-based diagnosis and treatment selection of $3,014, compared to $3,159 of CBPM (see Table D).
(f) Baseline cost for management of hypertension was estimated at $941/year when individuals are adherent to medication, and $567/year when they are not. These values are from MEPS (19) and corresponded to the mean expenditure per hypertensive adult among those receiving care for hypertension in 2010. We subtracted mean expenditure in prescription medications to assess management cost of non-adherent individuals. Costs were inflated to 2019 dollars using the medical inflation rate.  (20). Costs were inflated to 2019 dollars using the medical inflation rate.
(j) For the post-CVD states, we considered second-year cost of $11,162/year, $14,322/year and $5,630/year for MI, stroke and other CVD, respectively (20). Costs converge to baseline in year 5. Costs were inflated to 2019 dollars using the medical inflation rate.

Payer's benefits
From a commercial insurance perspective, the payer receives a premium revenue stream from all surviving members who remain in the insurance plan.
(o) Exiting the model (death or leaving the insurance plan) represents an elimination of the premium revenue stream. Average premiums were obtained from an average commercial insurance plan from eHealth website (21) and from the average Medicare advantage plan from CMS(22). Parameters are reported in Table E. Note that premium gains are independent of the cohort health status, and they are only affected by permanence in the plan. Since type of intervention does not affect insurance turnover rate, the only effect of SMBP on insurance premium savings is through its indirect effect on CVD mortality. Finally, commercial insurance companies reduce their financial burden through patient cost-sharing. In our model, we assume annual deductibles for health care services, as shown in Table E.

Sensitivity analysis
We performed sensitivity analyses to assess how the NPV of the SMBP investment changes with variations in the three SMBP use cases. With respect to diagnosis (Use Case #1), SMBP is known to have better diagnostic specificity than CBPM, but similar sensitivity (11). This enables SMBP to detect the white coat hypertension (WCH) and avoid false positive diagnoses. Results in Table F show that an investment in SMBP monitoring is sensitive to the accuracy of SMBP measurement in detecting WCH. A 10% reduction in SMBP specificity reduces the investment's NPV by more than 45%, with higher reductions among older age groups. For treatment selection (Use Case #2), we assume that physicians using CBPM require more visits to select and titrate treatment. Sensitivity analysis were performed to assess the effect of saving 2 visits with SMBP among those who had 4 or more visits with CBPM. We also assessed the effect of saving 1 visit among those with 3 or more visits, and among those with 5 or more visits. Results presented in Table G show a moderate impact on NPV, except for individuals aged 65 or older. For the ongoing BP monitoring (Use Case #3), we assume that SMBP improves the patient's medication adherence rate, which translates into more hypertensive patients (whether diagnosed by CMBP or SMBP) receiving treatment. We further assume that the use of SMBP monitoring reduces BP levels, with a reduced CVD risk in our Markov model. A sensitivity analysis was performed for both parameters (adherence to antihypertension medication and reduction in systolic BP). Results are presented in Table H. The standardized mean (s.m.) difference associated with greater medication adherence due to SMBP use is estimated at 0.21 (9), which represents an increment of nearly 1 percent in the percentage of individuals adherent to treatment compared to usual care. For example, for the age group 25-34 the percentage of individual adherent to medication treatment is estimated at 30.7% with an standard deviation of 4.4% based on NHANES(12) data. That puts the adherence rate under usual care at 30.7% and under SMBP at 31.7% (30.7% − 0.21 x 4.4%). For the sensitivity analysis we assessed the effect of a 10% increase and a 10% reduction in the baseline s.m. difference in medication adherence. Table H shows that the effect on NPV is not sensitive to changes in adherence, except for age group 75-84. A similar result is observed for the reduction in systolic BP associated to SMBP use. The impact on NPV is negligible for non-elderly adults.

Uncertainty analysis
We performed Monte Carlo simulations with 1,000 draws to account for uncertainty in parameters (probabilities, rates, costs and insurance cost sharing). For each age-group, we obtained the probability of a favorable NPV, assessed as the fraction of simulated Monte Carlo draws with a positive NPV. Table I shows the NPV distribution for each age group, indicating a more disperse distribution for older cohorts. Fig E presents the probability of a favorable overall NPV (NPV>0) by age group. We also show probabilities of a favorable NPV broken down by use case when SMBP is used only for diagnosis, only for treatment selection, or only for ongoing hypertension management. Results coincide with our main finding that diagnosis and treatment selection (Use Cases #1 and 2) drive the positive NPV of the SMBP investment. When SMBP is used only for BP monitoring (Use Case #3), the investment produces negative NPVs in most cases.