Baseline variables representing the year prior to the last statin supply dispensed before the patient screened positive were extracted from several sources. Prescription variables from the DSS National Data Extract included the total number of prescription fills, number of statin fills, and whether each subject received a prescription in each of 32 different classes of medications, including those used for atherosclerotic diseases (ischemic heart disease, etc) or risk factors (blood pressure, diabetes, etc) and depression. The VHA Medical SAS data files compiled by Austin Information Technology Center provided national information on VHA outpatient encounters and inpatient stays. All diagnoses recorded during the baseline year were used to identify chronic comorbidities based on an enhanced Elixhauser classification scheme.¹⁴ In addition, categories for ischemic heart disease (ICD-9-CM codes 410 through 414), heart failure (398.91, 402.01, 402.11, 402.91, 404.03, 404.11, 404.13, 404.91, 404.93, 425.4, 425.7-9, 428), and dementia (290.1-4, 290.9, 291.1-2, 294.1, 331.0-2, 331.82) were created. Monthly enrollment records were matched to the date of each subject’s last VHA statin supply to determine the duration of VHA enrollment; enrollment priority (1-8), which, in part, determines whether there was a VHA prescription copayment; and other active health insurance including Medicaid. Dates of death were obtained retroactively from monthly enrollment records.

Statistical Analyses

Descriptive statistics, mean ± standard deviation, median (25th-75th quartiles), and proportion (percentage) as appropriate for the level of measurement and distribution of the data are reported. Binomial 95% confi dence intervals are estimated for proportions. We did not collect follow-up data on patients who did not become positive screens, thus the sensitivity, specificity, and negative predictive value could not be estimated. Group characteristics were compared by Student t test for normally distributed continuous variables, Wilcoxon rank-sum test for skewed continuous measures, or Pearson’s χ² test for proportions. Reported P values are not adjusted for multiple comparisons. Stata software (version 10.1) was used for all analyses.

RESULTS

Within 5 months of beginning to screen the statin-recipient cohort approximately every 2 weeks, 1000 patients (4.6%; 95% confi dence interval [CI] 4.3%-4.8%) out of 21,935 became past due for a resupply of statin medication, ie, became a positive screen. Their statin prescriptions are characterized in Table 1. Consistent with the 120-day grace period and the predominant dispensing of a 90- day supply, the positive screens became past due 214 ± 23 days after their last statin supply was sent out. The positive screens received 3.1 ± 1.7 statin supplies and 249 ± 92 days of supply during the baseline year. Thus, the majority appeared to have stopped after receiving more than the required minimum of 2 statin supplies. Subsequently 824 in the statin user cohort of 21,935 (3.8%; 95% CI 3.5%-4.0%) were judged to be true positives based on not getting a resupply of any statin medication from the VHA for at least 4 months after their past supplies should have been consumed. The estimated positive predictive value was 824 out of 1000, or 82% (95% CI 80%-85%). During follow-up we discovered that 11% of the 824 true positives had died (n = 95) before their past-due date, or became nursing home residents (n = 17), where most likely they would not be personally responsible for taking or discontinuing the statin. Later DSS prescription extracts found 176 (0.8%; 95% CI 0.7%-0.9%) had received another supply before their calculated past-due date and therefore were false positives. The statin supplies defi ning false positive screens were all dispensed in February 2010, the first month of screening the prescription records. Characteristics of the false and true positives are compared in Table 2. There were no major differences in these or any other comorbidities or medication classes that are not shown, or variables listed in Table 1.

The follow-up survey was sent within 18 ± 10 days after the calculated past due date. Excluding the 95 deceased, ¹⁷ nursing home residents, and 21 cases with undeliverable addresses, the response rate was 786 out of 867 (91%), including 640 true and 146 false positives. Reasons reported by the true positives for no longer taking the statin provided by the VHA are listed in Table 3. Of the 640 true positives, 302 reported at least 1 reason that could be diffi cult for healthcare providers to address, including physician orders, side effects, switching to alternative treatments, or using non-VHA sources of supply. Other reasons that were given might be more amenable to efforts to reinstate long-term use of a statin. However, 208 of the remaining true positives that gave reasons that might be more amenable to reinstating use received another statin supply from the VHA within 6 months after becoming past due. Thus, efforts to reinstate their statin would not be required. The Figure summari zes the overall yield of good candidates for efforts by healthcare providers to reinstate long-term statin use.

DISCUSSION

The majority of patients identifi ed by this particular approach to screening electronic healthcare records were questionable candidates for reinstatement of long-term statin use. The yield could be improved in several ways. Having more up-to-date records of deaths and transitions into long-term institutional care would eliminate some that definitely did not require follow-up. A more advanced screening program that could determine whether VHA providers had discontinued the statin would be very helpful. Most likely this would require a search of textual notes. Searching for side effects such as elevated liver enzymes or a diagnosis of myopathy could exclude those with adverse effects that could have prompted discontinuation of a statin. However, some adverse effects, such as patients not liking how they felt on medication, will not be recorded or may be recorded as textual notes. Likewise, information on alternative treatments might not be readily accessible to an electronic screening program. Searching the healthcare record for cholesterol levels indicating adequate cholesterol control after patients became past due could further reduce the large number of positive screens needing follow-up.