PMCCPMCCPMCC

Search tips
Search criteria 

Advanced

 
Logo of nihpaAbout Author manuscriptsSubmit a manuscriptNIH Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
 
Arthritis Care Res (Hoboken). Author manuscript; available in PMC Dec 1, 2013.
Published in final edited form as:
PMCID: PMC3504186
NIHMSID: NIHMS385036
Does Medication Adherence Itself Confer Fracture Protection? An Investigation of the Healthy Adherer Effect in Observational Data
Jeffrey R Curtis, MD MS MPH,1 Huifeng Yun, MD, MSc,1,1 Jeff L. Lange, PhD,2 Robert Matthews,1 Pradeep Sharma, MS,1 Kenneth G. Saag, MD, MSc,1,1 and Elizabeth Delzell, ScD1
1University of Alabama at Birmingham, Birmingham, AL
2Procter & Gamble, Mason OH
Corresponding Author: Jeffrey R. Curtis, MD MS MPH 510 20th Street South Faculty Office Towers 805D Birmingham, AL 35294 ; jcurtis/at/uab.edu 205-975-2176
Background
Prior observational studies have shown an association between bisphosphonate adherence and fewer fractures. It is unclear if such studies reflect pharmacologic benefits or behavioral attributes, i.e. the healthy adherer effect.
Objective
To examine the association of therapy adherence and fracture risk among patients initiating therapies hypothesized to be favorable, unfavorable, or neutral toward fracture risk so as to evaluate for a healthy adherer effect.
Methods
In this observational study, we identified patients within Medicare 2006-09 data who initiated any of three medication groups within nine months after an osteoporotic fracture: 1) oral bisphosphonates (n = 2507), 2) selective serotonin reuptake inhibitors (SSRI) (n=2420), or 3) angiotensin converting enzyme inhibitor, or calcium channel blocker (ACE/CCB) (n=2178). Cox regression analysis, adjusting for covariates, was used to compare fracture rates at the hip and major osteoporotic fracture sites including hip, clinical vertebral, humerus, and wrist during follow-up comparing patients with high adherence versus low adherence within each medication group.
Results
There were few baseline differences between those who had high adherence versus lower adherence. High adherence with bisphosphonates decreased fracture risk at both hip (hazard ratio (HR) =0.53, 95% CI 0.32-0.96) and major fracture sites (HR = 0.61, 0.45-0.80). High adherence with SSRIs suggested increased fracture risk at both hip (HR = 1.58, 0.97-2.57) and major fracture sites (HR=1.32, 0.96-1.83). High adherence with ACE/CCBs was neutral toward fracture risk at both hip (HR = 1.27, 0.67-2.41) and major fracture sites (HR = 1.00, 0.67-1.49).
Conclusion
In this observational cohort of older individuals, the association between medication adherence and fracture risk differed by medication exposure, suggesting a limited role for the healthy adherer effect in observational studies of osteoporosis medications.
Keywords: osteoporosis, adherence, persistence, fracture, selective serotonin reuptake inhibitor, bisphosphonates, effectiveness
The efficacy of medical interventions is typically demonstrated by randomized controlled trials (RCTs) of carefully selected patients, usually with few comorbidities. The generation of evidence supporting the real-world effectiveness of those interventions can be challenging, and large and lengthy randomized controlled trials (RCTs) may not be feasible. Observational data provide such evidence but do not benefit from randomization, which in large samples, can be expected to balance all measured and unmeasured confounding factors. Analytic techniques such as propensity scores that can be used to control for confounding may not solve this problem since they balance only the factors that can be measured.
Among the potential threats to validity related to confounding in observational data is the so-called ‘healthy adherer’ effect. Similar in concept to the healthy worker effect (1), the concern is that patients who have high adherence to their medications may differ in many ways from patients who are prescribed the same medications but who choose not to adhere to them. While adherent and non-adherent patients may differ in many respects that can be measured (2), residual confounding may exist if there are important unmeasured factors (e.g. tobacco use, dietary factors including calcium and vitamin D intake, physical activity, obesity) related to medication adherence.
Empiric support for the healthy adherer effect comes largely from RCTs of patients with cardiovascular disease in which adherence to blinded placebo, given by the trial, has been shown to confer a significant mortality benefit(3, 4). The healthy adherer effect has also been suggested to exist amongst women participating in clinical trials of hormone therapy and bisphosphonate therapy, where the benefit of adherence with placebo was observed for a range of important health outcomes (5, 6). Important limitations of these studies were that patients were participants in a clinical trial, and adherence was to blinded study medication, not medications provided by patients’ physicians that were being used to treat known conditions. Additionally, potentially important confounders that could change over time were not able to be controlled for in a time-varying fashion in some of these analyses.
If a healthy adherer effect, similar to that observed in these trials, produced important confounding in observational data, this would compromise the validity of analyses that examine the relationship between medication adherence and health outcomes such as fracture. Indeed, multiple prior observational studies have shown high bisphosphonate adherence was associated with a significantly reduced fracture risk (7-9). The extent to which these types of analyses that quantify the association between osteoporosis medication adherence and fracture might be confounded by the healthy adherer effect in real-world settings is unclear. The healthy adherer effect might be especially problematic in circumstances where adherence impacts real-world effectiveness and if it differs between the medications being studied. We therefore evaluated the association between high adherence with different medications with varying effects on bone and fractures to provide evidence for whether the healthy adherer effect might affect observational analyses of real-world osteoporosis medication use.
Patient selection
Using a random 5% sample of enrollees in the United States Medicare program from 2006 to 2009, we identified individuals age >= 65 assumed to have osteoporosis on the basis of an incident hip, clinical vertebral, humerus, or wrist fracture. These fractures were defined as the ‘index fracture’. Eligible individuals must have had Medicare part A, part B in the 6 months prior to the index fracture and throughout follow-up. Part D pharmacy coverage was required in the 6 months prior to initiation of medications of interest (see below) and throughout follow-up. The age restriction was required since eligibility for coverage in the Medicare program for patients under age 65 is typically due to end stage renal disease requiring dialysis or medical disability; these younger patients with such conditions might have confounded the main associations of interest and introduce unwarranted heterogeneity and therefore were not included.
Medication exposure cohorts
Individuals who began oral bisphosphonates, selective serotonin reuptake inhibitors (SSRIs), and/or angiotensin converting enzyme inhibitors, or calcium channel blockers (ACE/CCB) in the 9 months subsequent to the fracture were identified (Figure 1). Eligibility required new use of these medications, defined by no use in the 6-month period prior to therapy initiation(10). These medications were chosen for their expected favorable, unfavorable, and no or minimal effect on fracture risk, respectively (11-17).
Figure 1
Figure 1
Study Timeline for Selecting Cohort
Patients initiating both oral bisphosphonates and ACE/CCBs in the 9 month period were assigned to the oral bisphosphonate cohort; patients initiating SSRIs and ACE/CCBs were assigned to the SSRI group. No other medication combinations were allowed. If patients used any non-allowed combination (e.g. bisphosphonate and SSRIs), they were excluded from analysis if this use occurred prior to the start of follow-up; they were censored if this occurred after follow-up had already begun.
Assessment of adherence and fracture outcomes
Adherence was quantified as the medication possession ratio [MPR](18) and was assessed 6 months after the initiation of a study medication. The purposes of this 6-month interval were 1) to allow for a minimally reasonable amount of time for the medications with effects on bone to exert their effects (19); 2) to allow enough time for the computation of MPR to be stable (20); and 3) to avoid classifying claims for follow-up care the index fracture as a second, incident fracture. Follow-up time for new fractures began at the time of this initial MPR measurement, and the MPR was updated every 90 days thereafter and analyzed in a time-varying fashion. Covariates other than MPR were assessed in the 12 months prior to the start of follow-up. A sensitivity analysis began observation time on the date of first use, rather than after a 6 month lag as in the primary analysis, to allow for comparability with external studies.
Fractures of interest included major osteoporotic fractures, defined as hip, clinical vertebral, humerus, and wrist/distal forearm(21). These were identified in the administrative data using published algorithms supported by validation studies that compared these algorithms to the gold standard of fractures confirmed through medical record review (22, 23). Incident fractures were identified using ICD-9 diagnosis codes and Healthcare Common Procedure Coding System (HCPCS) codes specific to the particular fracture site. Only “case qualifying” fractures were used in the analysis, which generally were defined as either an inpatient hospital claim with a discharge diagnosis of fracture, or a physician claim for fracture accompanied by a site-specific procedure code for surgical repair of the fracture, or a validated algorithm for clinical vertebral fractures. To discriminate between follow-up for a prior fracture and a new incident fracture at the same site, a period of least 90 days free of any claims for the same type of fracture was required for the patient to be allowed to be at-risk for a new fracture at the same site.
Statistical analysis
Adherence at 6 months was examined descriptively within each of the 3 cohorts. For analytic purposes, adherence was categorized as < 50% (low), 50-<80% (intermediate), >= 80% (high), following prior conventions (8, 24). Cox proportional hazard models evaluated the association between adherence outcomes within each medication group. For the major osteoporotic fracture outcome, the Cox models were stratified by the type of initial fracture. Censoring occurred at the time of first hip or major fracture; loss of Medicare part A, B, and D coverage; death; the end of the observation period (December 31st, 2009); or use of a disallowed medication combination (defined above). All analyses were conducted using SAS 9.2 (SAS Institute, Cary NC). The study was conducted with approval of the local institutional review board, and a data use agreement between the university and the Center for Medicare and Medicaid Services (CMS) governed use of the data. CMS policy prevented display of any cell sizes with counts less than 11.
We identified 2507 new users of oral bisphosphonates, 2420 new users of SSRIs, and 2178 new users of an ACE/CCB in the 9 months following a hip, clinical vertebral, humerus, or wrist/distal forearm fracture (Table 1). Approximately one-third to one-half of all initial fractures were hip fractures. Between three-quarters and ninety percent of all eligible individuals were women. The mean +− SD number of months of follow-up was similar across cohorts: 12.8 +− 9.2 months for the bisphosphonate cohort; 11.6 +− 8.9 months for the SSRI cohort; and 11.3 +− 8.9 months for the ACEI/CCB cohort. Within each medication cohort, there were not large differences in baseline covariates according to adherence measured at 6 months (Supplemental Table 1).
Table 1
Table 1
Characteristics of fracture patients initiating therapy, by type of medication
The unadjusted rates of major osteoporotic and hip fractures are shown for each of the three medication cohorts in Table 2, stratified by adherence category. The highest rates of both major osteoporotic fractures and hip fractures in the bisphosphonate cohort were observed among patients with MPR < 50%. Comparing the least adherent to the most adherent patients within each of the three medication cohorts, the crude rates of major osteoporotic fractures and hip fractures were lower among the adherent bisphosphonate users, higher among the adherent SSRI users, and comparable between the adherent and non-adherent ACE/CCB users.
Table 2
Table 2
Incidence rates of major osteoporotic fractures* and hip fractures by medication possession ratio category and type of medications
After multivariable adjustment for all factors listed in Table 1, highly adherent bisphosphonate users had a significantly reduced rate of major osteoporotic fractures compared to less adherent bisphosphonate users (adjusted hazard ratio (HR), 0.61; 95% confidence interval (CI), 0.45-0.80) (Figure 2). A similarly reduced rate of hip fractures among the highly adherent bisphosphonate users was observed (HR,0.53; 95% CI, 0.32-0.96). In contrast, the rate of both major osteoporotic fractures (HR, 1.32; 95% CI, 0.96-1.83) and hip fractures (HR, 1.58; 95% CI, 0.97-2.57) were numerically higher among those most adherent to SSRIs. There was no significant association between high adherence to ACE/CCBs and major osteoporotic fractures (HR, 1.00; 95% CI, 0.67-1.49) or hip fracture (HR, 1.27; 95% CI, 0.67-2.41). Considering other risk factors for fracture in these multivariable models, there were few other fracture risk factors besides medication adherence aside from age trends and diabetes (Table 3). The sensitivity analysis that began follow-up time on the date of the first prescription for each medication group of interest was generally concordant with the main analysis, although the protective associations between bisphosphonate adherence and fracture was attenuated towards the null (Supplemental Tables 2 and 3).
Figure 2
Figure 2
Adjusted* hazard ratios (HR) and 95% confidence interval (CI) for the association between medication adherence and major osteoporotic fractures** and hip fractures by type of medication
Table 3
Table 3
Adjusted* hazard ratio (HR) and 95% confidence interval (CI) for the association between bisphosphonate adherence, covariates of interest and fractures
Among this cohort of older U.S. patients with osteoporosis based upon the occurrence of a recent fracture, we observed that high medication adherence to bisphosphonates significantly decreased fracture risk. Consistent with known adverse effects of SSRIs on bone and fracture risk(14), high adherence to SSRIs significantly increased fracture risk, suggesting that if there was any healthy adherer bias present, it did not meaningfully impair our ability to detect the previously-described significant association between SSRI use and increased fracture risk. Finally, as was hypothesized in the absence of a healthy adherer effect, there was no association between adherence to ACE/CCBs and fractures; in fact, fracture rates were as high or slightly higher among patients with ACE/CCB MPRs >=80% than among those with MPRs <50%, although this difference was not statistically significant. As these medications have no known effect on bone or fracture risk, the lack of a protective association between high adherence to these medications and fracture provides reassurance that there was not an important healthy adherer bias present in this population.
Although some RCTs have suggested the existence of a healthy adherer bias for medications provided by a clinical trial (5, 6), this was not observed in this observational analysis of real-world data. While many factors are different between a trial setting and routine clinical care, patients’ and physicians’ knowledge of the indication for the medication might produce adherence patterns different from those seen for a medication provided in a blinded fashion and in accordance with a study protocol. Moreover, patients participating in trials may have different demographics, motivations, and comorbidities that limit the generalizability of results from trials to the real world. Finally, the duration of follow-up in the trials that have suggested a healthy adherer effect may be important if potential confounders could not be adjusted for in a time-varying fashion. The mean follow-up time in this analysis was relatively short, approximately 1 year, which limits the potential problem of misclassifying baseline confounders that could change over time.
The association between high medication adherence and other important factors and outcomes has been examined in a number of observational studies. High adherence to bisphosphonates has been shown to have a strong association with high adherence to other long term medications (20). High medication adherence also appears to be associated with a greater proclivity for health-seeking behaviors such as vaccination and cancer screening(25, 26). Providing some support for the existence of a healthy adherer bias, adherence to statin therapy has been shown to be associated with a lower likelihood of workplace accidents, motor vehicle crashes, and other effects unlikely to be attributable to statins (26, 27). However, evidence for a healthy adherer bias is not consistent in the literature. A study of patients that survived myocardial infarction found a mortality benefit among patients who adhered to statins and beta-blockers but not calcium channel blockers (28), suggesting the benefit of adherence was drug class specific and not an epiphenomenon of behaviors or other factors associated with adherence. Likewise, an osteoporosis-focused study found that high adherence with calcitonin and raloxifene did not confer any non-vertebral fracture benefit (29). Given that these medications have not been shown even in RCTs to provide meaningful non-vertebral fracture benefit, this result provides some reassurance for lack of an important healthy adherer bias in that population.
It was perhaps surprising that except for medication adherence, there were not strong associations between the risk factors that were examined and fracture. This was noted both in the similarity between the crude and adjusted hazard ratios for each multivariable model, as well as by the lack of significance of most risk factors as shown in Table 3. Indeed, aside from diabetes, income, and trends for older age, there were few additional risk factors for fracture. Although speculative, this may be due to the fact that the people in this age group (age 65+) with the major fractures required for eligibility in this analysis were relatively similar to one another in terms of frailty and other conditions, such that there were few other important fracture risk factors identifiable after controlling for medication adherence.
The strengths of our study include a large number of patients with osteoporosis defined by the occurrence of a recent fracture rather than on the basis of an administrative diagnosis code for osteoporosis, or even a bone mineral density test result. Our study design allowed us to assess adherence with three different medications that were expected to have favorable effects on fracture risk reduction (bisphosphonates), unfavorable effects on bone (SSRIs), and ACEI/CCBs that are not known to have important effects on bone. The benefit of examining the latter two medication groups is that we could evaluate adherence behaviors and associated factors with presumably less fracture-related channeling and confounding expected if only osteoporosis medications were able to be studied.
Despite these strengths, administrative data lack detailed clinical information on some risk factors for fracture including bone mineral density results, markers of frailty, and fall risk. However, administrative data have been shown to predict fracture risk comparably with clinical risk factors and the World Health Organization’s Fracture Risk Assessment Tool, FRAX®(30). Our population of patients included only individuals age >= 65, perhaps limiting the generalizability of our results to older patients. The number of fracture events was modest, especially for hip fractures; consequently, some analyses yielded wide confidence intervals. Finally, the cutpoints for categories of adherence selected from the literature may not have been optimal to discriminate between the fracture benefits of adherence with MPR of 50-<80% versus >=80%, as many of the patients in the 50-<80% category had MPR at the upper end of this range. Moreover, the dataset used did not capture patients who were prescribed but failed to ever fill the medications that were studied, so called ‘primary non-adherence’.
In conclusion, in this observational analysis of individuals enrolled in Medicare, we did not find strong evidence of a healthy adherer effect that suggested that medication adherence behaviors and associated factors conferred fracture benefit. Based upon this result, observational analyses evaluating the association between osteoporosis medication adherence and fracture risk in older patients may not be meaningfully confounded by a healthy adherer bias.
Significance & Innovation
  • The potential benefits of medication adherence in real-world settings are of increasing interest. A source of possible confounding in observational analysis is the ‘healthy adherer’ effect, whereby patients who adhere to medications may differ in important ways that affect outcomes compared to those who do not continue to take medications.
  • Among three cohorts of patients initiating medications with anticipated favorable (bisphosphonate), adverse (SSRI), and neutral (ACE or CCB) effects on bone, we found minimal evidence for a healthy adherer effect in this cohort of older patients with osteoporotic fractures
  • Based upon this result, observational studies of osteoporosis medication adherence and fracture risk may not be meaningfully confounded by the healthy adherer effect
Supplementary Material
Supp Table S1
Supp Table S2&S3
Acknowldegements
Dr. Curtis receives support from the NIH (AR 053351) and AHRQ (R01 HS018517). Dr. Saag receives salary support from the National Institutes of Health (AR052361) and the Agency for Healthcare Research and Quality (U18 HS016956). This research was supported by a contract between UAB and Amgen, Inc. Only the authors from UAB had access to the Medicare data used. The analysis, presentation and interpretation of the results were solely the responsibility of the authors.
1. Li CY, Sung FC. A review of the healthy worker effect in occupational epidemiology. Occup Med (Lond) 1999;49(4):225–9. [PubMed]
2. Brunner R, Dunbar-Jacob J, Leboff MS, Granek I, Bowen D, Snetselaar LG, et al. Predictors of adherence in the Women’s Health Initiative Calcium and Vitamin D Trial. Behav Med. 2009;34(4):145–55. [PMC free article] [PubMed]
3. Simpson SH, Eurich DT, Majumdar SR, Padwal RS, Tsuyuki RT, Varney J, et al. A meta-analysis of the association between adherence to drug therapy and mortality. BMJ. 2006;333(7557):15. [PMC free article] [PubMed]
4. Influence of adherence to treatment and response of cholesterol on mortality in the coronary drug project. N Engl J Med. 1980;303(18):1038–41. [PubMed]
5. Curtis JR, Larson JC, Delzell E, Brookhart MA, Cadarette SM, Chlebowski R, et al. Placebo adherence, clinical outcomes, and mortality in the women’s health initiative randomized hormone therapy trials. Med Care. 2011;49(5):427–35. [PubMed]
6. Curtis JR, Delzell E, Chen L, Black D, Ensrud K, Judd S, et al. The relationship between bisphosphonate adherence and fracture: Is it the behavior or the medication? results from the placebo arm of the fracture intervention trial. J Bone Miner Res. 2010 [PMC free article] [PubMed]
7. Imaz I, Zegarra P, Gonzalez-Enriquez J, Rubio B, Alcazar R, Amate JM. Poor bisphosphonate adherence for treatment of osteoporosis increases fracture risk: systematic review and meta-analysis. Osteoporos Int. 2010;21(11):1943–51. [PubMed]
8. Siris ES, Selby PL, Saag KG, Borgstrom F, Herings RM, Silverman SL. Impact of osteoporosis treatment adherence on fracture rates in North America and Europe. Am J Med. 2009;122(2 Suppl):S3–13. [PubMed]
9. Curtis JR, Westfall AO, Cheng H, Lyles K, Saag KG, Delzell E. The Benefit of Adherence with Bisphosphonates Depends on Age and Fracture Type: Results From an Analysis of 101,038 New Bisphosphonate Users. J Bone Miner Res. 2008;23(9):1435–41. [PMC free article] [PubMed]
10. Ray WA. Evaluating medication effects outside of clinical trials: new-user designs. Am J Epidemiol. 2003;158(9):915–20. [PubMed]
11. Liu B, Anderson G, Mittmann N, To T, T. A, N. S. Use of selective serotonin receptor uptake inhibitors or tricyclic antidepressants and risk of hip fractures in elderly people. Lancet. 1998;351:1303–7. [PubMed]
12. Diem SJ, Blackwell TL, Stone KL, Yaffe K, Haney EM, Bliziotes MM, et al. Use of antidepressants and rates of hip bone loss in older women: the study of osteoporotic fractures. Arch Intern Med. 2007;167(12):1240–5. [PubMed]
13. MacLean C, Newberry S, Maglione M, McMahon M, Ranganath V, Suttorp M, et al. Systematic review: comparative effectiveness of treatments to prevent fractures in men and women with low bone density or osteoporosis. Ann Intern Med. 2008;148(3):197–213. [PubMed]
14. Wu Q, Bencaz AF, Hentz JG, Crowell MD. Selective serotonin reuptake inhibitor treatment and risk of fractures: a meta-analysis of cohort and case-control studies. Osteoporos Int. 2012;23(1):365–75. [PubMed]
15. Solomon DH, Mogun H, Garneau K, Fischer MA. Risk of fractures in older adults using antihypertensive medications. J Bone Miner Res. 2011;26(7):1561–7. [PubMed]
16. Gagne JJ, Patrick AR, Mogun H, Solomon DH. Antidepressants and fracture risk in older adults: a comparative safety analysis. Clin Pharmacol Ther. 2011;89(6):880–7. [PubMed]
17. Rejnmark L, Vestergaard P, Mosekilde L. Treatment with beta-blockers, ACE inhibitors, and calcium-channel blockers is associated with a reduced fracture risk: a nationwide case-control study. J Hypertens. 2006;24(3):581–9. [PubMed]
18. Sikka R, Xia F, Aubert RE. Estimating medication persistency using administrative claims data. Am J Manag Care. 2005;11(7):449–57. [PubMed]
19. Harrington JT, Ste-Marie LG, Brandi ML, Civitelli R, Fardellone P, Grauer A, et al. Risedronate rapidly reduces the risk for nonvertebral fractures in women with postmenopausal osteoporosis. Calcif Tissue Int. 2004;74(2):129–35. [PubMed]
20. Curtis JR, Xi J, Westfall AO, Cheng H, Lyles K, Saag KG, et al. Improving the prediction of medication compliance: the example of bisphosphonates for osteoporosis. Med Care. 2009;47(3):334–41. [PMC free article] [PubMed]
21. [Accessed May 31st, 2008]; http://shef.ac.uk/FRAX.
22. Ray WA, Griffin MR, Fought RL, Adams ML. Identification of fractures from computerized Medicare files. J Clin Epidemiol. 1992;45(7):703–14. [PubMed]
23. Curtis JR, Mudano AS, Solomon DH, Xi J, Melton ME, Saag KG. Identification and validation of vertebral compression fractures using administrative claims data. Med Care. 2009;47(1):69–72. [PMC free article] [PubMed]
24. Siris ES, Pasquale MK, Wang Y, Watts NB. Estimating bisphosphonate use and fracture reduction among US women aged 45 years and older, 2001-2008. J Bone Miner Res. 2011;26(1):3–11. [PubMed]
25. Brookhart MA, Patrick AR, Dormuth C, Avorn J, Shrank W, Cadarette SM, et al. Adherence to lipid-lowering therapy and the use of preventive health services: an investigation of the healthy user effect. Am J Epidemiol. 2007;166(3):348–54. [PubMed]
26. Patrick AR, Shrank WH, Glynn RJ, Solomon DH, Dormuth CR, Avorn J, et al. The association between statin use and outcomes potentially attributable to an unhealthy lifestyle in older adults. Value Health. 2011;14(4):513–20. [PubMed]
27. Dormuth CR, Patrick AR, Shrank WH, Wright JM, Glynn RJ, Sutherland J, et al. Statin adherence and risk of accidents: a cautionary tale. Circulation. 2009;119(15):2051–7. [PMC free article] [PubMed]
28. Rasmussen JN, Chong A, Alter DA. Relationship between adherence to evidence-based pharmacotherapy and long-term mortality after acute myocardial infarction. JAMA. 2007;297(2):177–86. [PubMed]
29. Cadarette SM, Solomon DH, Katz JN, Patrick AR, Brookhart MA. Adherence to osteoporosis drugs and fracture prevention: no evidence of healthy adherer bias in a frail cohort of seniors. Osteoporos Int. 2011;22(3):943–54. [PMC free article] [PubMed]
30. Yun H, Delzell E, Ensrud KE, Kilgore ML, Becker D, Morrisey MA, et al. Predicting hip and major osteoporotic fractures using administrative data. Arch Intern Med. 2010;170(21):1940–2. [PubMed]