A systematic review of published work on adherence of oral medications for diabetes reported that 15–33 percent of patients failed to have adequate adherence for oral agent doses as prescribed (Cramer 2004
). Other meta-analytic and systematic reviews suggest similar estimates that are primarily based on secondary adherence measures (Brown et al. 1999
; Donnan, MacDonald, and Morris 2002
; Ray 2003
; Cramer 2004
; Odegard and Capoccia 2007
;). In our cohort, 28.9 percent (95 percent CI ± 0.55 percent; 5,422/18,770) were classified as having inadequate adherence (≥20 percent gap in medication) based on CMG, a conventional measure of secondary adherence, while 47.4 percent (95 percent CI ± 0.59 percent; 12,954/27,329) received that classification based on NPMG. Thus, our findings suggest that the current public health burden associated with inadequate medication adherence may be larger than previously thought. Poor adherence among ongoing users was low (2.7 percent), but failure to initiate or obtain even one refill is a substantial problem (22.3 percent).
A substantially poorer medication adherence rate is observed when using a new prescription cohort, and accounting for those who fail to initiate the new medication (primary nonadherence), fail to ever refill (early nonpersistence), and time after discontinuation, rather than the more commonplace approach of only observing ongoing users. Because researchers were previously incapable of constructing large, new prescription cohorts, the public health burden attributable to poor medication adherence has been significantly underappreciated. The new prescription cohort design has the added advantage of enabling researchers to distinguish failure to initiate therapy (primary nonadherence) from the failure to prescribe appropriate therapy (clinical inertia), as well as facilitating better accounting of medication stockpiling (which can bias adherence estimates).
Although Kaiser Permanente physicians would likely have obtained a patient's agreement about starting new medication before writing a prescription, 3–9 percent of the patients prescribed new cardiometabolic therapies never filled the new prescription. In one of the few large studies of primary adherence, Canadian researchers evaluated primary adherence for cardiac and noncardiac discharge medications prescribed for secondary prevention after hospitalization for an acute myocardial infarction (Jackevicius, Li, and Tu 2008
). These authors reported similar rates of primary nonadherence for antihypertensives (e.g., 3.8 percent for ACE inhibitors) and cholesterol-lowering medications (e.g., 5.2 percent for statins) as in the present study, although higher primary nonadherence rates for diabetes medications (14.6 percent for insulin and 13.3 percent for oral agents). A study in another managed care organization (Harvard Vanguard Medical Associates) recently reported low primary nonadherence rates for oral diabetes medications (under 10 percent) (Trinacty et al. 2009
), which are closer to our observations of 4 percent. Differences may be attributable to the case mix and type of prescriptions. The Kaiser sample included prevalent diabetes patients, whereas the Harvard sample included only new onset diabetes cases. However, both groups focused on ambulatory prescriptions, whereas the Canadian study included prescriptions first issued during a hospitalization. Primary nonadherence may reflect inadequate communication or shared decision making (failure of the provider to assess patient readiness to initiate a new therapy and gain a patient's buy-in) (Heisler et al. 2007
). It could also reflect a lack of trust in the provider or unaddressed patient concerns (e.g., side effects, inability to pay).
While few studies have evaluated primary nonadherence, several studies have evaluated early nonpersistence (i.e., filling once, but never refilling) in new user cohorts. In this study, 16–22 percent of the cohort obtained the initial dispensing but never refilled despite easy access to a pharmacy (every Kaiser medical facility has an onsite pharmacy and a mail order pharmacy service is available). While one earlier study of diabetic patients (1997–2000), suggested lower rates of early nonpersistence for antihyperglycemics (Hertz, Unger, and Lustik 2005
), most studies are consistent with ours in reporting high rates of early discontinuation of oral diabetes medications (Trinacty et al. 2009
), statins (Avorn et al. 1998
; Benner et al. 2002
; Jackevicius, Mamdani, and Tu 2002
;), and antihypertensives (Elliott et al. 2007
; Siegel, Lopez, and Meier 2007
;). Even when cardiometabolic medications were prescribed at discharge following hospitalizations for myocardial infarction (Prospective Registry Evaluating Myocardial Infarction Study or PREMIER), the largest decline in medication use was observed within 1 month of discharge (one in five discontinued use of β
-blockers or statins); thereafter medication use remained relatively stable (Ho et al. 2006b
). These findings are qualitatively consistent with our findings.
Estimates from this study are not readily compared with the existing literature given the differing methodologies used. The one-fifth of diabetes patients who never become ongoing users of newly prescribed cardiometabolic medications would be excluded from summary measures of secondary adherence. Moreover, our ability to censor patient follow-up when their discontinuation was clinically recognized (34 percent of total new prescriptions) rather than misclassifying it as a form of nonadherence likely explains the low (4 percent) rate of discontinuation at later stages (“second-stage nonpersistence,” i.e., discontinuation after the initial two fills and during the 24-month follow-up).
The validation study suggests that, among ongoing users, NPMG performed similarly to a well-accepted, objective secondary adherence measure (CMG) in predicting response to newly initiated statin therapy. Moreover, NPMG showed a significant relationship to self-reported adherence and out-of-pocket costs, while CMG did not, although the differences between the measures were not clinically relevant. Findings suggest that NPMG is a valid objective, summary adherence measure and, because it is numerically identical to CMG for persistent users, recommend it as a viable alternative when the necessary data are available.
Pharmacy-based measures have limitations. While medications gaps, strictly speaking, indicate pharmacy underutilization, they provide strong evidence of nonadherence; patients cannot take medications that are not dispensed. However, these measures of adherence average gaps over the entire follow-up and we are not able to pinpoint when the nonadherence occurred. Moreover, we are unable to verify whether patients actually took dispensed medications; therefore, levels of nonadherence estimated from pharmacy utilization data must be viewed as lower bounds. While we set generous time limits to allow dispensings to occur (e.g., remaining days supply plus 90 days grace period) to identify discontinuation, dispensings may occur after follow-up ended for a few, exceedingly nonadherent patients. We have conducted a sensitivity analysis and found that only~3 percent of patients had a subsequent refill after we determined them to have discontinued based on our decision rule. We did search electronic medical records for evidence that a physician initiated a stop medication order or prescribed an alternative medication as an indication that ending therapy (during the same time period) was clinically recognized. However, attributing the remaining discontinuations as strictly a patient decision may somewhat overestimate discontinuation by the patients without apparent consultation with their provider. Finally, because both CMG and NPMG integrate adherence over an extended observation window for their respective populations (ongoing users and all patients prescribed a new medication, respectively), they are imperfect predictors of clinical effectiveness. Ideally, the timing of the adherence assessment should immediately precede the follow-up clinical measurements for response to therapy initiation (e.g., postinitiation LDL measure). Unfortunately, stockpiling and variations in timing of refills make it difficult, if not intractable, to reliably estimate adherence for short periods proximal to the clinical outcome of interest. For that reason, self-report at the time of the clinical outcome measure might be the most relevant adherence measure despite the potential shortcomings (e.g., social desirability and white coat compliance) for the purpose of linking adherence to clinical response.
We cannot entirely rule out some pharmacy utilization at non-Kaiser pharmacies, which is not captured in our databases. Kaiser members with pharmacy benefits have a financial incentive to use only Kaiser pharmacies because the benefits are not honored at non-Kaiser pharmacies. In 2005–2006, we evaluated survey responses from 20,188 Kaiser diabetic patients involved in the NIDDK-funded Diabetes Study of Northern California
) (Moffet et al. 2009
) regarding their out-of-plan pharmacy utilization in the previous 12 months. Of the 96 percent who had a pharmacy benefit, non-Kaiser pharmacies were used less than a single time (0.4 times) during the previous year. Because we excluded the 4 percent lacking prescription benefits, the underascertainment of pharmacy utilization in this present study should be minimal.
In some settings, pharmaceutical distributors provide free samples to physicians. Patients receiving such free samples might try the medication but quit before ever obtaining the first recorded dispensing at the pharmacy. However, pharmaceutical representatives have no direct access to Kaiser providers, who are not permitted to receive or provide patients with free samples. We excluded patients who had a hospitalization during follow-up to avoid misclassifying early nonpersistence as primary nonadherence, for patients never refilling a new prescription order issued after discharge, when an initial course of a new medication was administered in the hospital.
NPMG provides a valid, comprehensive measure of adherence, which can be used in addition to existing metrics currently available to study medication adherence. Secondary adherence measures (e.g., CMG, as a proxy for current adherence) and self-reported recent adherence are the most appropriate exposures to estimate adherence for currently persistent patients. The advantages of using NPMG come into play when it is used as an outcome measure because it more comprehensively summarizes overall adherence among patients newly prescribed a medication, without excluding particularly nonadherent subgroups. NPMG should be an appropriate tool for the evaluation of the effectiveness of population-level interventions aiming to improve adherence, or summarize overall adherence to a clinical trial medication protocol. It may help explain lower than expected effectiveness of population-based pharmacotherapy interventions (e.g., prescribing all high-risk patients the polypill; Reddy 2007
). By capturing initiation and early adherence, NPMG may be sensitive to disadvantageous characteristics of a new medication, which may be barriers to ongoing use such as high out-of-pocket costs, poor side effect profile, and low patient acceptability. NPMG may also be useful as a quality improvement measure or as a feedback measure for providers or at the health care facility level. While not appropriate for linkage to clinical outcomes when medication taking has only short-term effects (e.g., blood pressure), NPMG may be useful to establish an integrated measure of total drug exposure (to adjust prescribed drug dosages) in studies of adverse drug effects with long lag times (e.g., cancer risk with past drug exposures). While efficient monitoring of primary nonadherence may require electronic prescribing data (not yet available at most health care facilities), early nonpersistence can be identified using pharmacy utilization data; NPMG can be easily modified for use among primary adherent cohorts (new user cohorts) as long as it is understood that it slightly overestimates total adherence because it will ignore primary nonadherence. Finally, although this study used cardiometabolic therapies in diabetes to illustrate the methods, our findings likely have relevance for adherence in other chronic diseases.
The need for therapy intensification reoccurs over the course of diabetes (United Kingdom Prospective Diabetes Study [UKPDS] 1995
) and other chronic diseases, and thus the failure of patients to become persistent users may occur repeatedly, with cumulative negative health consequences. Using electronic prescribing records, we can now capture potential nonadherence associated with each intensification attempt, estimate adherence in a more comprehensive way, and intervene in a timely fashion in the patients who fail to become ongoing users. A prescription for a new cardiometabolic medication implies a long-term commitment to therapy, and providers need to recognize that there may be considerable reluctance on the part of patients to make such a commitment. Interventions to address adherence need to be multifaceted (technical, behavioral, and educational) and sensitive to the myriad causes for such patient resistance, including provider, medication-specific, and system-level factors (Pound et al. 2005
; Odegard and Capoccia 2007
; Polonsky 2007
; Huang et al. 2009
;). The relative importance of these factors likely differs across the stages of medication taking; thus, barriers and promoters of adherence need to be evaluated separately for each stage of adherence. A surprising finding in this study was that the proportion of newly prescribed patients that never became ongoing users (n
=6,098; 22 percent) was eightfold greater than the proportion who maintained ongoing use, but with inadequate adherence (n
=744; 3 percent). Moreover, observed rates of discontinuation at the later stages of medication taking were much lower than previously reported and likely due to our ability to censor follow-up at the point a provider stopped or switched the patient's medication. While public health efforts have primarily focused on improving adherence and reducing nonpersistence among ongoing users, clearly more attention is needed to address these issues at the earliest stages after a new medication is prescribed.