記住我
Patient failure to adhere to prescribed medication regimens is a persistent and serious problem in the U.S. health care system.1 The literature on this topic shows significant medication nonadherence in a wide range of patient populations, with an average nonadherence rate nationally of 24.8% and rates in some subpopulations as high as 95%.2,3 Numerous behavioral, social, economic, medical, and policy-related factors have been identified as contributing to medication nonadherence.4 Medication nonadherence is associated with increased hospitalization rates and emergency department visits, higher mortality rates, worse patient outcomes, and increased downstream costs, all of which impose heavy, avoidable health care costs on society.5–10 Despite efforts to increase drug insurance coverage and improve both physician-pharmacist-patient communication and patient education, nonadherence rates among patient populations remain high as successful interventions are generally substantially complex and costly.11–13
Cost-related medication nonadherence (CRN) is a particular challenge. A recent study shows that one in 4 adults in the United States has a difficult time affording their medications.14 There is a growing body of evidence in recent years that even after the implementation of Medicare part D, CRN has persisted, or even worsened, among the sickest Medicare patients, including those with multiple chronic conditions, depression, and stroke survivors, according to cross-sectional comparisons at different time points.11,15,16
CRN is an important dimension in patient-centered outcomes research, which is defined as “the evaluation of questions and outcomes meaningful and important to patients and caregivers.”17 Patient-centered outcomes research focuses attention on patients’ beliefs, preferences, and needs, and places an emphasis on shared decision-making by patients and physicians.18,19 CRN, when present, reflects the misalignment of unmet individual patient needs and physician-prescribed medical interventions, and CRN behaviors may take place outside physicians’ knowledge. In the presence of CRN, shared decision-making may not be possible as the foundation for such decision-making is broken down. From a policy perspective, CRN resides in the conjunction of individualized care and barriers due to systematic factors and individual characteristics. Hence, understanding the prevalence and predictors of CRN provides policymakers with a barometer of disfranchisement among patients and the pathways to remedy such disfranchisement.
Although there is an accumulated body of literature on CRN, such literature consists predominantly of cross-sectional studies using yearly surveys. Such studies fail to distinguish those with persistent CRN from transient and intermittent CRN, and transient or intermittent CRN may have patient health impacts and social costs different from persistent nonadherence. Hence, given limited resources, those demonstrating persistent CRN should be prioritized in medical and policy interventions, as they suffer the most from the consequences of CRN and possibly create the largest downstream economic losses. A metric that differentiates persistent CRN from transient and intermittent CRN was recently developed,20 but its performance in integrated datasets that use multileveled CRN analysis, include those who were deceased, and have a longer follow-up is unknown. There are a few nationally representative surveys that carry questionnaires on CRN, such as the Medicare Current Beneficiary Survey, the Health and Retirement Survey, the National Health Interview Survey, and the National Financial Capability Study.21–24 Among these, however, cross-sectional design prevents the development of longitudinal measures of persistent CRN behaviors, biennial design makes it difficult to ascertain the detailed gravity of patients’ CRN behavior beyond a general generational trend,25 and rotating panel design significantly limits a survey’s utility in studying the longitudinal aspects of CRN.
We, therefore, developed an integrated dataset of CRN by following up on a sample of 2000 predominantly Black Medicare patients at high risk of hospitalization, cared for at an urban academic medical center, on a quarterly basis for 8 quarters. This patient population has been a policy focus for their high needs and high-resource utilization, but their daily struggles in paying for medications and other basic needs are not well known due to the paucity of data. This study aimed to advance the literature in 3 ways: (1) integrating the survey data with Medicare enrollment and claims data to evaluate the level of CRN and the risk factors for its elevation, (2) achieving complete capturing of the CRN characteristics of the high-cost, high-need patient population by evaluating those who were deceased during the study period, and (3) extending the follow-up period from 1 year to 2 years to evaluate the longer-term tenacity of CRN. Medicare patients have high-resource utilization during their last year of life,26–28 and their struggles with receiving care and maintaining dignity are important research topics for patient-centered outcomes research.
PATIENTS AND METHODSWe used a sample of 2000 Medicare patients at high risk of hospitalization who had enrolled in a study of the Comprehensive Care Physician (CCP) Model conducted in an urban academic medical center between November 2012 and June 2016, during which time the economy was in steady growth with no major shock.29 To qualify for this care model, the patients needed to have been hospitalized at least once in the past year or in the care of the emergency department when enrolled in the study. Our internal analysis indicated that this enrollment criterion selected patients at 300%–400% of the average annual health care expenditure for Medicare beneficiaries during the study’s base year, and 86% of the patients had Part D coverage when enrolled. The high-need, high-resource utilization patient population represented by this sample needs urgent policy intervention.21 Working with this population allows us to study the longitudinal aspects of CRN for those at heightened risk for it. The original CCP study is a randomized trial with 2 arms of intervention and control; for this study, we pooled the 2 arms together and controlled in the regression analysis to reflect the population average and treat the study arm similarly as an enabling factor.20
The patients enrolled in the CCP study were surveyed every 3 months about CRN behavior based on 4 questions adapted from the survey questionnaires of the Medicare Current Beneficiary Survey. Patients were asked whether, during the past 3 months, they had ever done the following due to cost: (1) not filled or refilled a prescription, (2) delayed filling a prescription, (3) skipped doses, or (4) taken smaller doses to make medication last longer.22 CRN then was categorized as 1 if the patients reported any of these 4, and 0 if none. Patients who did not answer CRN questionnaires but completed the follow-up survey for the quarter during the 8-quarter follow-up period were treated as not reporting CRN for that quarter.
We linked the survey data to the Medicare Master Beneficiary Summary File, Vital Statistics File (VSF), and Part A and B claims data through the Health Insurance Claim (HIC) number reported by patients during study enrollment, as well as HICs recorded in our electronic health records (EHR). Procedures were performed to cross-check the HICs in these 2 data sources to reduce entrance errors during the surveys. These HICs were sent in a Finder File to the contractor of the Centers for Medicare and Medicaid Services for linking to the Medicare data. When the linked data returned, we performed a linkage integrity check to ensure the birthdates of patients identified in the Medicare database and in our survey database matched.
The Medicare VSF contains patient death dates up to the month when the VSF was generated. Our goal is to establish 2-year mortality after enrollment. However, because the Centers for Medicare and Medicaid Services uses multiple sources of data, including claims, date of death submitted by a patient’s family member, and benefits information used to administer the Medicare program collected from the Railroad Retirement Board and the Social Security Administration,30 there is concern about the accuracy and completeness of VSF death information. We hence devised a multistep process to harmonize the mortality information using multiple data sources for cross-validation.
Among Medicare patients, there are about 12.5 million people with Medicare-Medicaid dual eligibility who are at peril of lack of care coordination. These patients fall at the lowest rung of the economic ladder (hence are qualified for Medicaid) and are either elderly or disabled and have been a focus of policy to improve care coordination in recent years.31 How this important subpopulation fares in CRN has clear policy significance but is not well understood. For those dual eligibles, the opposing forces of health insurance (enabling) and low individual income (disabling) work to offset each other, and it is imperative for practitioners and policymakers to understand the net outcome of those offsetting effects on individual patients and other factors which may be conducive for increasing or decreasing CRN. We hence developed a baseline measurement of Medicare-Medicaid dual eligibility by looking back 1 year before study enrollment using Master Beneficiary Summary File and classifying as dual-eligible those patients who were so eligible during the year before enrollment. This approach aims to reduce bias due to churning in dual status,32 as patients may be caught in the revolving door of dual eligibility. Medicare Part D enrollment was also controlled for in the regression analysis.
We developed a CRN metric describing persistent, intermittent, and transient CRN during the 2-year follow-up, with persistent CRN (more than 50% of the quarterly assessments), intermittent CRN (more than 20% but ≤50% of assessments), and transient CRN (at least once but ≤20% of assessments).20 Such a classification system constitutes a classic ordinal scale, as the order is known but not the distance between values.33 Accordingly, we developed an ordered logit model to evaluate risk factors including dual eligibility, socio-demographic variables (including age, sex, race, ethnicity, and educational attainment), health variables [including subjective general health perception and deficiencies in Activities of Daily Living34 and Instrumental Activities of Daily Living,35 and total Medicare costs, including Part A and B in the year before enrollment to the study. Age was rescaled by 10 years to better show the age effect. For Medicare total costs, including Part A and B, we scaled by quartiles to reduce noise and better show the relative cost quartiles for this sample of high-cost, high-resource utilization patients. The race variable was self-reported, considered the gold standard in research.36 The proportional odds assumption was tested using an approximate likelihood-ratio test in the ordered logit model to ensure its validity. We obtained predictive margins for probabilities by age for the categorical CRN to reflect the effect of age on the population level, and the odds ratios (ORs) from the regression analyses to reflect the adjusted effect of individual factors as CRN concerns both policy-level instrument and individual factors. This CRN metric only counted completed surveys in the denominator over the 8-quarter study period; hence the bias due to intermittent missingness or dropouts was mitigated. The main driver for dropouts was mortality, and we controlled for a range of health variables in our regression analysis. As persistent CRN is of particular concern to medical practice and health policy, we further developed a logit model to assess the effect of the set of variables described above on this aspect of CRN behaviors. We performed a sensitivity analysis for categorical CRN using multinomial logistic regression with the same set of covariates. The analysis was conducted using Stata Statistical Software version 14 (StataCorp LP).
RESULTSAn initial sample of 2000 patients was recruited for the CCP study. We excluded 8 subjects due to duplicate allocation during the randomization process, and 4 subjects eventually withdrew from the study. For the remaining 1988 subjects, 12 were not linked to Medicare data due to missing HIC or erroneous HIC entries in our database. The remaining 1976 (99.4%) subjects were ascertained for accurate linkage by matching the birthdates between our EHR/survey datasets and birthdates in the VSF. We further excluded 60 subjects who were in Medicare Advantage plans when enrolled due to missing cost information for the year before enrollment in the study, resulting in a sample size of 1916 before harmonizing mortality among various data sources.
Through reviewing our EHRs and Medicare claims data, we concluded that 365 deaths reported in the Medicare VSF were likely true as there were no more clinical care/claims activities after the death dates. Further cross-referencing the deaths by using National Death Index, we discovered 3 deaths recorded in our EHRs/National Death Index but not reflected in the Medicare VSF, making a total of 368 deaths during the 2-year follow-up. We harmonized the death and death dates of both EHR and VSF-indicated deaths during the follow-up period and concluded that the VSF death dates were more likely to be accurate to the true death dates based on clinical activity, which is consistent with the very small percent of nonvalidated deaths in VSF in recent years.25 With the harmonized death data, we were able to exclude 87 subjects who were deceased during the first 90 days after enrollment to the CCP study and thus did not reach the point of the first follow-up survey, and 68 patients who survived the first 90 days but did not complete any follow-up surveys, among whom 29 were deceased during the 2-year follow-up, resulting in the final sample size of 1761. The final number of deaths among the 1761 subjects was 252 (14.3%).
Table 1 shows the socio-demographic and health characteristics of the 1761 subjects by categorical CRN. Overall, the mean age was 62.6 years (SD, 16.6), 1099 (62%) were women, 1527 (87%) were Black. 412 (23%) had educational attainment of less than high school, 1152 (65%) were dual-eligible, 67 (4%) rated their general health as excellent, 753 (43%) had one or more deficiencies in ADL, 1132 (64%) had deficiencies in Instrumental ADL, 474 (27%) reported depression, and mean total Medicare costs in the year before enrollment were $39,900 (SD, $52,600). General health perception and depression were associated with categorical CRN (P < 0.01, for both).
TABLE 1 - Socio-Demographic and Health Characteristics of the Study Sample Full sample (N = 1761); N (%) No CRN (N = 892); N (%) Transient CRN (N = 296); N (%) Intermittent CRN (N = 395); N (%) Persistent CRN (N = 178); N (%) P Age (mean, SD) 62.6 (16.1) 65.2 (16.4) 62.4 (16.3)* 60.1 (15.0)* 56.5 (14.5)* — Femele (reference: male) 1099 (62) 553 (62) 169 (57) 265 (67) 112 (63) 0.06 Race Black 1527 (87) 765 (86) 258 (87) 351 (89) 153 (86) 0.79 White 106 (6) 60 (7) 16 (5) 18 (5) 12 (7) — Other race 128 (7) 67 (8) 22 (7) 26 (7) 13 (3) — Hispanic ethnicity 67 (4) 38 (4) 10 (3) 13 (3) 6 (3) 0.79 Medicare-Medicaid dual eligibility 1152 (65) 601 (67) 200 (68) 241 (61) 110 (62) 0.09 Educational attainment Junior high school or less 129 (7) 69 (8) 19 (6) 31 (8) 10 (6) — Some high school (grades 9–11) 283 (16) 152 (17) 55 (19) 54 (14) 22 (12) — High school graduate (grade 12) 505 (29) 255 (29) 90 (31) 107 (27) 53 (30) — Some college or junior college 514 (29) 241 (27) 81 (27) 126 (32) 66 (37) 0.20 College graduate 176 (10) 92 (10) 29 (10) 35 (9) 20 (11) — Postgraduate 97 (5) 48 (5) 14 (5) 30 (8) 5 (3) — Do not know 35 (2) 24 (3) 3 (1) 7 (2) 1 (1) — Refused 21 (1) 11 (1) 4 (1) 5 (1) 1 (1) — General health perception: Excellent 67 (4) 46 (5) 7 (2) 9 (2) 5 (3) <0.01 Very good 160 (9) 74 (8) 36 (12) 38 (10) 12 (7) — Good 446 (25) 247 (28) 79 (27) 94 (24) 26 (15) — Fair 676 (38) 330 (37) 113 (38) 159 (40) 74 (42) — Poor 409 (23) 193 (22) 60 (20) 95 (24) 61 (34) — Do not know 3 (<1) 2 (<1) 1 (<1) 0 0 — Deficiency in ADLs 0 1008 (57) 502 (56) 179 (60) 229 (58) 98 (55) 0.62 1 311 (18) 156 (17) 54 (18) 72 (18) 29 (16) — 2 126 (7) 61 (7) 18 (6) 33 (8) 14 (8) — 3 or more 316 (18) 173 (19) 45 (15) 61 (15) 37 (21) — Deficiency in IADLs 0 629 (36) 318 (36) 114 (39) 142 (36) 55 (31) 0.10 1 177 (10) 93 (10) 31 (10) 39 (10) 14 (8) — 2 149 (8) 59 (7) 32 (11) 36 (9) 22 (12) — 3 or more 806 (46) 422 (47) 119 (40) 178 (45) 87 (49) — Depression (reference: no depression) 474 (27) 198 (22) 78 (26) 128 (32) 70 (39) <0.01 Medicare costs prior year, $1000s (mean, SD) 39.9 (52.6) 43.3 (56.4) 33.4 (42.2)* 36.6 (48.8)* 41.3 (55.7) — CCP intervention 885 (50) 464 (52) 130 (44) 196 (55) 95 (53) 0.09 Part D enrollment 1514 (86) 767 (86) 266 (90) 324 (82) 157 (88) 0.02*Statistically significant at P <0.05 when compared with those with no CRN in t tests.
P values were derived from the χ2 tests.
ADL indicates Activities of Daily Living; CCP, comprehensive care physician; CRN, cost-related medication nonadherence; IADL, Instrumental Activities of Daily Living.
Among the 1761 subjects, 1083 (61.5%) completed all 8 quarterly surveys. The 252 patients who were deceased constituted 37% of those not completing all surveys, with intermittent missingness during the study period accounting for the rest. The mean number of completed follow-up surveys was 6.6 (SD, 2.2). Two hundred fifty-two (14.3%) subjects were deceased during the 2-year follow-up. Among these 1761 patients, 968 (55%) reported CRN at least once in the 24-month study period. Using a metric of persistent, intermittent, and transient CRN,20 only 187 (11%) reported persistent CRN, 453 (25.7%) reported intermittent CRN, and 328 (18.6%) reported transient CRN. In other words, among those who reported CRN, 19.3% reported persistent CRN.
Figure 1 shows the observed overall CRN rates by age group, which showed higher CRN rates among the younger patients. Figure 2 shows the predictive margins of probabilities of reporting CRN categories by age groups at the population level in the ordered logit model. There was a clear pattern of declining probabilities of persistent and transient CRN with increasing age, while transient CRN was flat.
FIGURE 1: Observed CRN prevalence by age. Ages were grouped by 10; those at the ages of 18 and 19 were combined with those in their 20s, and those 100 or older were combined with those in their 90s. CRN indicates cost-related medication nonadherence.
FIGURE 2: Predictive margins of probabilities by age for categorical CRN: results from the ordered logit model. Ages were grouped by 10; those at the ages of 18 and 19 were combined with those in their 20s, and those 100 or older were combined with those in their 90s. CRN indicates cost-related medication nonadherence.
Table 2 shows the data sources that constituted the integrated analytic data. Table 3 shows the conditional effect of factors on categorical CRN and persistent CRN from the ordered logit model and single logistic regression, respectively. Dual eligibility seemed to be protective in both equations [adjusted OR (AOR) = 0.53, P < 0.01; AOR = 0.45, P < 0.01, respectively], and depression was associated with reporting higher levels of categorical CRN and persistent CRN (AOR = 1.53, P < 0.01; AOR = 1.55, P = 0.01, respectively). Lower general health perception was in general associated with reporting a higher level of categorical CRN (P < 0.01) but was not statistically significantly associated with persistent CRN.
TABLE 2 - Data Sources for Variables in the Integrated Dataset Type of variable Variable Data source Note Outcome CRN (transient, intermittent, and persistent) Quarterly follow-up surveys up to 8 quarters CRN was categorized by a metric based on the frequency (intensity) of reporting CRN during the 8-quarter follow-up Variable for sample construction Baseline Medicare managed care enrollment Medicare MBSF Study sample was restricted to those starting with traditional Medicare as the baseline costs were unknown otherwise Explanatory variables Age, sex, race, ethnicity, and education attainment Intake surveys at baseline Self-reported race is the gold standard Explanatory variable Dual eligibility MBSF One-year look-back before enrollment to reduce status churning Explanatory variable Part D enrollment MBSF Part D enrollment during the calendar year of enrollment to the study Explanatory variable Total Medicare costs Medicare Part A and B claims Summation of all claims during the 365 d before enrollment in the study Explanatory variables Depression, general health perception, deficiencies in ADLs, and IADLs Intake surveys at baseline Deficiencies in ADLs and IADLs are summed up with 0, 1, 2, 3, or moreADL indicates Activities of Daily Living; CRN, cost-related medication non-adherence; IADL, instrumental Activities of Daily Living; MBSF, Master Beneficiary Summary File.
留言 (0)