This study compared physicians’ perceptions of patients’ adherence to oral second-generation antipsychotics with the patients’ actual medication usage as determined from pharmacy claims. Our approach was different from those of other studies assessing physician perceptions of adherence. Previous studies assessing adherence to oral antipsychotics have compared physician estimates of adherence with patient self-reports, pill counts, electronic monitoring, electronic refill information and blood monitoring (22
). In a review of antipsychotic adherence research, the most commonly used method for assessing adherence was patient self-report (25
). Electronic refill records were used in only a small percentage of studies (8 of 161) and pharmacy claims were not used in any study (25
). In studies where physicians estimated adherence for individual patients, the assessment period was short, the patient population was small and drawn from a convenience sample, and it was unclear whether the physicians and patients had an established long-term relationship (22
). In contrast, our study asked physicians to estimate adherence of specific patients in their practice whom they had seen regularly for at least 1 year. The hypothesis was that familiarity with each patient’s history and behaviours would enable physicians to make more informed assessments of treatment adherence. In addition, research in patients receiving second-generation antipsychotics revealed that patients who had previous antipsychotic prescriptions were more likely to follow their medication regimen for a longer period than those receiving antipsychotics for the first time (27
). Together, these elements—experienced physicians who have an established relationship with patients who are accustomed to their antipsychotic regimen—represent a best-case scenario and promise a high degree of accuracy in physicians’ ability to estimate individual adherence rates.
Our results, however, support previous findings of physician overestimation of adherence (4
). Despite having treated the patients for at least 1 year, the surveyed physicians indicated approximately three-quarters of their patients were highly adherent to their therapeutic regimen, while claims-based data showed that fewer than half of the patients were highly adherent. The physicians also classified as highly adherent 82 (72%) of 114 patients who were found to have low-to-moderate claims-based adherence. Neither formal training in adherence nor the experience level of the physicians improved the physicians’ ability to estimate performance.
Discrepancies between physician perceptions of adherence and patient perspectives have been studied in many chronic illnesses, including mental health conditions (4
). Our results provide further support for the established finding that physicians overestimate adherence, whether the physicians’ estimates are based on higher level, general assessments of patients with a particular disease state, a moderate level of specificity, such as the patients in a general practice population, or, as our study found, at the individual patient level (4
). Familiarity with the patient did not improve the ability of the physicians to estimate treatment adherence. Previous research on physician estimates of patient adherence, whether in mental health or other therapeutic areas, was directed to higher levels of assessment and, to our knowledge, our study is novel in comparing physician estimates of adherence of individual patients with claims data. A similar design was used by Copher et al. in their study of adherence among patients with osteoporosis; however, although physicians were surveyed about patients in their own practises, the assessments were kept at the practice level rather than the patient level (28
To improve adherence, researchers have called for treatment teams to try to better understand the reasons patients fail to take medications (32
). That strategy presupposes the ability of physicians to recognise non-adherence in their patients. Our results show that physicians have difficulty estimating adherence, even in patients they see regularly.
The study had several limitations. Patients and their physicians were identified from a large administrative claims database with claims from employer-sponsored health plans in the US. The results from this study may not be generalisable to other physician and patient populations because of the relatively small number of physicians responding to the survey and the patients about whom they were asked. The physicians who took part in this study were asked about high functioning patients that were covered by an employer’s health plan and their responses may not be generalisable to other populations including patients with public health insurance or no insurance. Likewise, the patients in the study population may not be generalisable to other patients with mental disorders, or to patients in countries other than the US with different systems of health care. The majority of patients with schizophrenia in the US do not have commercial health insurance; usually they are covered by public health insurance or they are uninsured (33
). The patients with schizophrenia included in this study were covered by their own employers’ health plans or the plans of their spouses. This suggests these patients were functioning at a level that enabled them to maintain a relationship or steady employment.
Administrative claims data may contain diagnostic or treatment coding errors, and although there was a record of prescriptions filled, there was no way to determine whether the patients actually took the medication as prescribed and likely represents an underestimation of adherence.The patients also could have received medication samples or had prescriptions filled by pharmacies outside the health plans captured in the HIRD. Although many antipsychotics have FDA indications for bipolar disorder, they are not necessarily recommended for or prescribed to all individuals with bipolar disorder. When evaluating antipsychotic adherence using the MPR in individuals with bipolar disorder, we could not distinguish between gaps in treatment that were because of non-adherence vs. gaps in treatment or complete discontinuation that were clinically indicated. Finally, physicians were asked to estimate adherence over a 12-month period which may be subject to recall bias whereas the claims-based estimates of adherence were based on prescription fill data.