Improved compliance measures: Applications in an ambulatory hypertensive drug trial To assess the value of improved monitoring of medication-taking behavior in a drug trial, we employed a modified pill vial with microcircuitry to record the precise times when the vials were opened. After a 3-week placebo washout period, 21 ambulatory subjects with mild hypertension (mean age, 57 years; 67% men; 76% white) randomly received isradipine or enalapril twice daily in a double-blind titration during 10 weeks. Both drugs achieved a 13% reduction in sitting diastolic blood pressure (p < 0.01) with minimal symptomatic or laboratory toxicity. Although pill counts indicated near-perfect compliance (92% to 99% for both groups), the electronic monitor showed that fewer than half of all openings occurred at the prescribed interval of 12 -1- 2 hours. Modest overdispensing was documented in the 3 days before scheduled visits. The monitor confirmed that pill count misdassified compliance sufficiency in 22% of visits and permitted more discrete attribution for drug-associated adverse reactions and secondary resistance to treatment. We conclude that the electronic monitor reduces ambiguity about medication compliance and helps interpret both the biology and pharmacology of the trial. (CLIN PHARMACOL THER 1990;48:676-85.)
Peter Rudd, MD, Shaheda Ahmed, MD, Valerie Zachary, RN, Carol Barton, RN, MS, and Delphine Bonduelle, MD Stanford, Calif. The classic drug trial seeks an unambiguous comparison of two or more regimens, sometimes including inactive placebo treatment. Most commonly the primary focus is on demonstrating therapeutic efficacy and phannacologic safety. At each return visit the clinicianinvestigator must interpret the clinical experiment. If the patient achieves the therapeutic goal, the clinicianinvestigator retains the regimen; if not, the regimen may change. This simple algorithm assumes near-perfect medication compliance. Compliance itself is usually defined as the extent to which the patient adheres to the clinician's prescription.' Until recently, the most available
and feasible methods to assess compliance included patient self-report, pill counts, and biologic assays. Despite their well-documented imperfections ,2,3 these methods allow some quantification of medicationtaking behavior. Distinct from biologic or pharmacologic factors, the behavioral component may dramatically affect the clinical result. This report describes the application of a new technology to improve assessment of dynamic medicationtaking behavior within a drug trial. It quantifies the magnitude of some distortions introduced by relying on traditional measures (self-report and pill count) compared with information generated by an electronic medication monitor.
From the Department of Medicine, Stanford University Medical Center. Supported in part by Sandoz Pharmaceuticals, East Hanover, N.J. Presented in part at the Compliance Monitoring Symposium (Center for the Study of Drug Development, Tufts University, Iselin, N.J., Nov. 11, 1988) and the Annual Meeting of the American Society for Clinical Pharmacology and Therapeutics, Nashville, Tenn., March 9, 1989. Received for publication May 31, 1990; accepted Sept. 4, 1990. Reprint requests: Peter Rudd, MD, Division of General Internal Medicine, Room X-2I6 MSOB, Stanford University Medical Center, Stanford, CA 94305-5475.
METHODS Study design. The study consisted of a randomized, double-blind, parallel comparison of two medications among ambulatory patients with mild essential hypertension: isradipine (a dihydropyridine calcium channel blocker)49 and enalapril (a nonsulfhydryl angiotensinconverting enzyme inhibitor). l0-14 After a 3-week, single-blind placebo washout phase, we randomized qualified subjects into two treatment arms, comparing oral administration of the two drugs administered twice daily, titrated upward during the first 6 of 10 weeks on
13/1/25246
676
VOLUME 48 NUMBER 6
active treatment. The current report reflects results at one of six collaborating sites with the same protocol but the only site using electronic monitoring of compliance. Inclusion/exclusion criteria. Men or nonpregnant, nonlactating women 18 years of age or older either previously treated for or newly diagnosed with essential hypertension were included in the study. All subjects signed informed consent in accordance with institutional guidelines and were instructed that the dispensing vials containing their medication were "modified to allow follow-up on how the medications were taken." No other mention was made of monitoring compliance or the possibility of exclusion from the study for suboptimal compliance. To progress to randomization and active antihypertensive treatment, subjects had to (1) discontinue all previous antihypertensive medications by the initial visit, (2) exhibit average sitting diastolic blood pressure of 95 mm Hg or greater and 105 mm Hg or lower at the last two placebo washout visits, and (3) display 80% or greater compliance by pill count during the placebo treatment phase. We excluded individuals who failed to meet inclusion criteria or who gave evidence of secondary or malignant hypertension, recent onset of unstable cardiovascular disease requiring concomitant medication, other major chronic illness including psychiatric disorders, recent use of investigational or toxic medications, history of alcohol or other substance abuse, known hypersensitivity to study drugs or related compounds, evidence of autoimmune disease, abnormal laboratory test results for gastrointestinal, hepatic, or renal function, or intercurrent use of other confounding drugs. The study protocol further specified that subjects should (1) receive their study drugs before breakfast and before supper during all study phases; (2) be instructed to take the morning dose on the day of each clinic evaluation visit; (3) be considered noncompliant and be discontinued and replaced if an average of six or more doses per 2-week interval were missed during the first 6 weeks of the double-blind, active treatment period (6 doses /28 possible doses or 21.4% noncompliance); and (4) be considered noncompliant and data excluded from the statistical analysis for efficacy if more than 20% of the prescribed active doses were missed during the entire study. Medication regimen. The medication titration procedure included four levels for each drug, corresponding respectively to 1.25, 2.5, 3.75, and 5.0 mg for isradipine or 2.5, 5.0, 10.0, or 20.0 mg for enalapril, each at 12-hour intervals. The regimen dosage was escalated in double-blind fashion if the average of two
Monitoring medication compliance 677 diastolic determination in the sitting position remained greater than 90 mm Hg, as defined by Korotkoff fifth phase. Once the "goal" diastolic blood pressure in the sitting position was achieved, the regimen dosage was fixed for the remainder of the study. The dosage for the final two return visits remained constant, and no dosage reductions were permitted. Medication monitoring. At each visit subjects received their prescribed medications in a modified No. 4 amber, 30-dram plastic pill vial whose top contained a microprocessor capable of recording the precise time and duration of each vial opening (MEMS-1 [Medication Event Monitoring System]; Aprex Corporation, Fremont, Calif.). Internal circuitry permitted exclusion of multiple openings separated by 2 or fewer seconds and contained sufficient memory reserve to record up to 350 consecutive opening events during several months. Both the manufacturer's and our own validation studies confirmed near-perfect concordance (r = 0.9998) between interopening intervals by investigator log and by medication monitor during 2 to 8-week trials with 0.1 to 130-hour intervals. Subjects received a supply of pills corresponding to 150% of the prescription for each intervisit interval. At each return visit we reissued the same monitor to each subject with a new supply of pills and counted the remaining pills after the patient's departure. After study completion, we downloaded the monitor for subsequent data analysis by microcomputer. We acknowledged that patients might easily misrepresent their compliance by taking out multiple pills at one time and even discarding them rather than taking them at regular intervals as prescribed. In anticipation, we urged subjects to keep all study medications in the special vials and to remove only one pill at a time for immediate consumption. As a validation, we scanned the vial opening recording patterns for systematic deviations, such as taking out all medications for the two daily doses at one time. We assumed it was improbable that subjects would systematically open and close the vial at regular intervals for several weeks and yet fail to ingest the pill. Accordingly, the absence of any recorded openings during an interval likely indicated nonconsumption of the drug. Compliance measures. At each return visit we asked patients, "When did you take your last study pill," noting the specific self-reported time to (1) establish the probability of full drug effect from the morning's dose and (2) assess the accuracy of self-report compared with the monitor. We calculated compliance by four measures: (1) pill count, [number of missing pills/number of pills prescribed for the interval] x 100; (2) total
CLIN PHARMACOL THER DECEMBER 1990
678 Rudd et al.
140
COMPLIANCE BY MEMS MONITOR (%)
120
ioo 80 60
40 20 0
0
20
40 60 80 100 COMPLIANCE BY PILL COUNT (%)
120
140
Fig. 1. Concordance in compliance measures: pill count versus electronic monitor at 1- to 2-week intervals. Distribution of values for simultaneous measurement of medication-taking behavior among 18 hypertensive subjects followed up for 13 weeks, comparing pill counts with electronic medication monitors, which recorded each pill vial opening (n = 18 at four to seven visits for each subject).
openings, [total opening per intervisit interval/ total number of pills prescribed for the interval] x 100; (3) two-dose days, [number of days in which two dosings occurred by the monitor/ number of days in the interval] x 100; and (4) optimal intervals, [number of interdose intervals corresponding to the near-optimal 10 to 14 hours/total number of vial openings for the interval] x 100. We chose a stringent criterion for "optimal" dosing to simulate clinical situations in which adverse consequences may occur with major deviations from the prescribed regimen. Statistical analyses. Statistical analyses included descriptive statistics, X2 with Yates' correction for continuity with categoric data, and the t test statistics, correlation, and linear regression for continuous variables. The upper limit of statistical significance was p = 0.05.
RESULTS Patient characteristics. A total of 24 subjects entered the trial, but 3 failed to reach randomization (two did not meet diastolic blood pressure entry criteria, and one had intolerable fatigue with placebo). The remaining 21 subjects who completed the trial represented predominantly middle-aged patients with chronic, stable, mild hypertension with a tendency for obesity. Their mean age ( ± SD) was 56.9 ± 10.6 years, with an average duration of hypertension of 8.2 years (range, 0.5
to 21 years). Two thirds of the patients were men, 76% were white, 19% were black, and 5% were AsianAmerican. There were no systematic differences in
medication-taking behavior between the treatment groups, so the remainder of this report lumps all subjects together to highlight their overall patterns. Blood pressure response. By traditional measures, each of the drugs successfully and significantly lowered mean blood pressure from baseline (p .
24
INTEROPENING INTERVAL (hours) Fig. 2. Interopening intervals by monitor for regimen prescribed every 12 hours. Distribution of intervals between vial openings as measured by the MEMS monitor among 18 subjects; less than half of the openings occurred at the prescribed interdose interval for either drug and 25% of openings occurred less than 6 or more than 18 hours between dosings.
91.5% ± 1.6%, and isradipine, 98.7% .± 2.3% (difference not significant). For most subjects the compliance rates by pill count for 1- to 2-week intervals clustered at 100% ± 20%, but the range extended from 0% to 135%.
Monitor-generated compliance measures. Only 18 subjects generated a complete set of monitor-generated data because of prototype malfunction. We observed only moderate concordance of monitor vial opening data with pill count for each intervisit interval, as shown in Fig. 1. Fig. 1 plots the correlation between medication by pill count and total openings, as defined previously. Not all patients had functional medication monitors for all visits; we plotted the data only from intervisit intervals when both pill count and total opening data by monitor were available. Although the majority of points cluster along the line of identity (r = 0.243; p < 0.02), the scatterplot confirms that most outliers corresponded to low compliance by monitor despite near-perfect compliance by pill count. The most deviant subject exhibited a mean compliance by pill count of 93% for eight return visits during 89 days, whereas his monitor confirmed only 21% appropriate openings for the intervals. In fact the subject failed to open the medication vial on 65% of the days. Six of
18 subjects exhibited no vial openings on at least 10% of the monitored days (subgroup mean for percent of monitored days with no openings, 27%; range, 10% to 65%), although their compliance rates by pill count averaged more than 94%. Major deviation in medication-taking behavior would have remained unsuspected without the monitor's data. Compliance distributions. In part, the degree of apparent deviation from optimal compliance depends on the criteria to the study data, the mean values fall progressively below the optimal level of 100%, and the distributions by medication subgroup become more dissimilar. Table I summarizes the mean values for each treatment group by four definitions; with small sample sizes, none of the intergroup differences is significant at the p < 0.05 level, although all MEMS-based rates are significantly lower than those by pill count. Fig. 2 displays the distribution of interdose intervals by monitor-generated data. Fewer than half of all openings occurred at the prescribed interval of 12 -± 2 hours. Even if one extends the acceptable range to 6 to 18 hours, about 25% of openings took place outside
this generous range.
Impact of scheduled visits. The approach of the scheduled visit occasionally prompted unrepresentative
680 Rudd et
CLIN PHARMACOL THER DECEMBER 1990
al.
DOSES: Last 3 Days'/Whole Interval's (%)
120 100
80 60
113
observations r p
g
0.731 0.001
40 20 0 0
20
40
DAYS:
60
80
100
120
Last 3/Whole Interval (%)
Fig. 3. Homogeneity of drug taking: impact of scheduled visits. Correlation between doses taken (vial openings) and days passed. If scheduled visits stimulate atypical medication-taking behavior, one would expect that the percentage formed by the last 3 days compared with the entire interval between visits would not correlate with the percentage of doses taken in the last 3 days compared with those taken during the entire intervisit interval. Monitor data: n = 18, four to eight visits each.
Table I. Comparison of compliance measures Compliance measure Pill counts (all visits) (%) MEMS monitor (%) Total openings Two openings! day Openings at 10-14-hr intervals
Enalapril
lsradipine
All subjects
1.6*
98.7 ± 2.31-
95.3
2.1t
84.8 6.0t 57.4 ± 6.4t 34.4 ± 4.4t
95.6 ± 2.9§ 69.1 ± 4.9§ 43.1 ± 4.5§
89.5 ± 62.6 ± 38.3 ±
4.411 3.311
91.5
3.8M
Data are mean values -± SEM. Intergroup differences: none significant at 0.05 level. Intragroup differences: all significant at p 0.05. *n = 11; in = 10; in = 21; In = 8; in = 18.
behavior. For some subjects, atypical if near-perfect compliance or apparent overdispensing might occur to make up for previous deficiencies. If so, the traditional pill count would miss the discrepancies, because over dispensing could compensate for previous underdispensing. Fig. 3 exhibits the distribution of matching days with doses. If concordance were perfect, one would expect the number of days between scheduled visits to be proportional to the number of vial openings during the interval. In search of steady-state conditions, we
concentrated on the 3 days before each scheduled return visit, roughly corresponding to five times the half-life of the shorter-acting medication. We then plotted the percentage formed by the 3 days before the scheduled visit to the entire interval in days versus the percentage of openings for the same ratio of times. For example, on the abscissa, 3 days would equal 20% of a 15-day intervisit interval. During the same 3 days, we should expect that 20% of the doses should be dispensed and consumed.
VOLUME 48 NUMBER 6
Monitoring medication compliance TIME OF DAY (24 hr 0100k)
CASE
681
#406
41111.1.11.11 vertigo
2000 1600
1200 -
800
400
3/20
I
3/24
'
i
I
3/28
I
4/1
DATE
I
4/5 VISIT
Fig. 4. Patient-reported vertigo after "hypercompliance" with enalapril with monitor-generated data. Data from medication taking of subject 406, who reported 1 week of vertigo at his return visit on April 7, 1988. Each dot represents one medication vial opening corresponding to one 2.5 mg dosing. On the day before the onset of symptoms, the monitor showed seven openings rather than the usual one to three per day. The symptom, in turn, prompted less medication taking.
Overall, the correlation is high (r = 0.731; p
Peter Rudd, MD, Shaheda Ahmed, MD, Valerie Zachary, RN, Carol Barton, RN, MS, and Delphine Bonduelle, MD Stanford, Calif. The classic drug trial seeks an unambiguous comparison of two or more regimens, sometimes including inactive placebo treatment. Most commonly the primary focus is on demonstrating therapeutic efficacy and phannacologic safety. At each return visit the clinicianinvestigator must interpret the clinical experiment. If the patient achieves the therapeutic goal, the clinicianinvestigator retains the regimen; if not, the regimen may change. This simple algorithm assumes near-perfect medication compliance. Compliance itself is usually defined as the extent to which the patient adheres to the clinician's prescription.' Until recently, the most available
and feasible methods to assess compliance included patient self-report, pill counts, and biologic assays. Despite their well-documented imperfections ,2,3 these methods allow some quantification of medicationtaking behavior. Distinct from biologic or pharmacologic factors, the behavioral component may dramatically affect the clinical result. This report describes the application of a new technology to improve assessment of dynamic medicationtaking behavior within a drug trial. It quantifies the magnitude of some distortions introduced by relying on traditional measures (self-report and pill count) compared with information generated by an electronic medication monitor.
From the Department of Medicine, Stanford University Medical Center. Supported in part by Sandoz Pharmaceuticals, East Hanover, N.J. Presented in part at the Compliance Monitoring Symposium (Center for the Study of Drug Development, Tufts University, Iselin, N.J., Nov. 11, 1988) and the Annual Meeting of the American Society for Clinical Pharmacology and Therapeutics, Nashville, Tenn., March 9, 1989. Received for publication May 31, 1990; accepted Sept. 4, 1990. Reprint requests: Peter Rudd, MD, Division of General Internal Medicine, Room X-2I6 MSOB, Stanford University Medical Center, Stanford, CA 94305-5475.
METHODS Study design. The study consisted of a randomized, double-blind, parallel comparison of two medications among ambulatory patients with mild essential hypertension: isradipine (a dihydropyridine calcium channel blocker)49 and enalapril (a nonsulfhydryl angiotensinconverting enzyme inhibitor). l0-14 After a 3-week, single-blind placebo washout phase, we randomized qualified subjects into two treatment arms, comparing oral administration of the two drugs administered twice daily, titrated upward during the first 6 of 10 weeks on
13/1/25246
676
VOLUME 48 NUMBER 6
active treatment. The current report reflects results at one of six collaborating sites with the same protocol but the only site using electronic monitoring of compliance. Inclusion/exclusion criteria. Men or nonpregnant, nonlactating women 18 years of age or older either previously treated for or newly diagnosed with essential hypertension were included in the study. All subjects signed informed consent in accordance with institutional guidelines and were instructed that the dispensing vials containing their medication were "modified to allow follow-up on how the medications were taken." No other mention was made of monitoring compliance or the possibility of exclusion from the study for suboptimal compliance. To progress to randomization and active antihypertensive treatment, subjects had to (1) discontinue all previous antihypertensive medications by the initial visit, (2) exhibit average sitting diastolic blood pressure of 95 mm Hg or greater and 105 mm Hg or lower at the last two placebo washout visits, and (3) display 80% or greater compliance by pill count during the placebo treatment phase. We excluded individuals who failed to meet inclusion criteria or who gave evidence of secondary or malignant hypertension, recent onset of unstable cardiovascular disease requiring concomitant medication, other major chronic illness including psychiatric disorders, recent use of investigational or toxic medications, history of alcohol or other substance abuse, known hypersensitivity to study drugs or related compounds, evidence of autoimmune disease, abnormal laboratory test results for gastrointestinal, hepatic, or renal function, or intercurrent use of other confounding drugs. The study protocol further specified that subjects should (1) receive their study drugs before breakfast and before supper during all study phases; (2) be instructed to take the morning dose on the day of each clinic evaluation visit; (3) be considered noncompliant and be discontinued and replaced if an average of six or more doses per 2-week interval were missed during the first 6 weeks of the double-blind, active treatment period (6 doses /28 possible doses or 21.4% noncompliance); and (4) be considered noncompliant and data excluded from the statistical analysis for efficacy if more than 20% of the prescribed active doses were missed during the entire study. Medication regimen. The medication titration procedure included four levels for each drug, corresponding respectively to 1.25, 2.5, 3.75, and 5.0 mg for isradipine or 2.5, 5.0, 10.0, or 20.0 mg for enalapril, each at 12-hour intervals. The regimen dosage was escalated in double-blind fashion if the average of two
Monitoring medication compliance 677 diastolic determination in the sitting position remained greater than 90 mm Hg, as defined by Korotkoff fifth phase. Once the "goal" diastolic blood pressure in the sitting position was achieved, the regimen dosage was fixed for the remainder of the study. The dosage for the final two return visits remained constant, and no dosage reductions were permitted. Medication monitoring. At each visit subjects received their prescribed medications in a modified No. 4 amber, 30-dram plastic pill vial whose top contained a microprocessor capable of recording the precise time and duration of each vial opening (MEMS-1 [Medication Event Monitoring System]; Aprex Corporation, Fremont, Calif.). Internal circuitry permitted exclusion of multiple openings separated by 2 or fewer seconds and contained sufficient memory reserve to record up to 350 consecutive opening events during several months. Both the manufacturer's and our own validation studies confirmed near-perfect concordance (r = 0.9998) between interopening intervals by investigator log and by medication monitor during 2 to 8-week trials with 0.1 to 130-hour intervals. Subjects received a supply of pills corresponding to 150% of the prescription for each intervisit interval. At each return visit we reissued the same monitor to each subject with a new supply of pills and counted the remaining pills after the patient's departure. After study completion, we downloaded the monitor for subsequent data analysis by microcomputer. We acknowledged that patients might easily misrepresent their compliance by taking out multiple pills at one time and even discarding them rather than taking them at regular intervals as prescribed. In anticipation, we urged subjects to keep all study medications in the special vials and to remove only one pill at a time for immediate consumption. As a validation, we scanned the vial opening recording patterns for systematic deviations, such as taking out all medications for the two daily doses at one time. We assumed it was improbable that subjects would systematically open and close the vial at regular intervals for several weeks and yet fail to ingest the pill. Accordingly, the absence of any recorded openings during an interval likely indicated nonconsumption of the drug. Compliance measures. At each return visit we asked patients, "When did you take your last study pill," noting the specific self-reported time to (1) establish the probability of full drug effect from the morning's dose and (2) assess the accuracy of self-report compared with the monitor. We calculated compliance by four measures: (1) pill count, [number of missing pills/number of pills prescribed for the interval] x 100; (2) total
CLIN PHARMACOL THER DECEMBER 1990
678 Rudd et al.
140
COMPLIANCE BY MEMS MONITOR (%)
120
ioo 80 60
40 20 0
0
20
40 60 80 100 COMPLIANCE BY PILL COUNT (%)
120
140
Fig. 1. Concordance in compliance measures: pill count versus electronic monitor at 1- to 2-week intervals. Distribution of values for simultaneous measurement of medication-taking behavior among 18 hypertensive subjects followed up for 13 weeks, comparing pill counts with electronic medication monitors, which recorded each pill vial opening (n = 18 at four to seven visits for each subject).
openings, [total opening per intervisit interval/ total number of pills prescribed for the interval] x 100; (3) two-dose days, [number of days in which two dosings occurred by the monitor/ number of days in the interval] x 100; and (4) optimal intervals, [number of interdose intervals corresponding to the near-optimal 10 to 14 hours/total number of vial openings for the interval] x 100. We chose a stringent criterion for "optimal" dosing to simulate clinical situations in which adverse consequences may occur with major deviations from the prescribed regimen. Statistical analyses. Statistical analyses included descriptive statistics, X2 with Yates' correction for continuity with categoric data, and the t test statistics, correlation, and linear regression for continuous variables. The upper limit of statistical significance was p = 0.05.
RESULTS Patient characteristics. A total of 24 subjects entered the trial, but 3 failed to reach randomization (two did not meet diastolic blood pressure entry criteria, and one had intolerable fatigue with placebo). The remaining 21 subjects who completed the trial represented predominantly middle-aged patients with chronic, stable, mild hypertension with a tendency for obesity. Their mean age ( ± SD) was 56.9 ± 10.6 years, with an average duration of hypertension of 8.2 years (range, 0.5
to 21 years). Two thirds of the patients were men, 76% were white, 19% were black, and 5% were AsianAmerican. There were no systematic differences in
medication-taking behavior between the treatment groups, so the remainder of this report lumps all subjects together to highlight their overall patterns. Blood pressure response. By traditional measures, each of the drugs successfully and significantly lowered mean blood pressure from baseline (p .
24
INTEROPENING INTERVAL (hours) Fig. 2. Interopening intervals by monitor for regimen prescribed every 12 hours. Distribution of intervals between vial openings as measured by the MEMS monitor among 18 subjects; less than half of the openings occurred at the prescribed interdose interval for either drug and 25% of openings occurred less than 6 or more than 18 hours between dosings.
91.5% ± 1.6%, and isradipine, 98.7% .± 2.3% (difference not significant). For most subjects the compliance rates by pill count for 1- to 2-week intervals clustered at 100% ± 20%, but the range extended from 0% to 135%.
Monitor-generated compliance measures. Only 18 subjects generated a complete set of monitor-generated data because of prototype malfunction. We observed only moderate concordance of monitor vial opening data with pill count for each intervisit interval, as shown in Fig. 1. Fig. 1 plots the correlation between medication by pill count and total openings, as defined previously. Not all patients had functional medication monitors for all visits; we plotted the data only from intervisit intervals when both pill count and total opening data by monitor were available. Although the majority of points cluster along the line of identity (r = 0.243; p < 0.02), the scatterplot confirms that most outliers corresponded to low compliance by monitor despite near-perfect compliance by pill count. The most deviant subject exhibited a mean compliance by pill count of 93% for eight return visits during 89 days, whereas his monitor confirmed only 21% appropriate openings for the intervals. In fact the subject failed to open the medication vial on 65% of the days. Six of
18 subjects exhibited no vial openings on at least 10% of the monitored days (subgroup mean for percent of monitored days with no openings, 27%; range, 10% to 65%), although their compliance rates by pill count averaged more than 94%. Major deviation in medication-taking behavior would have remained unsuspected without the monitor's data. Compliance distributions. In part, the degree of apparent deviation from optimal compliance depends on the criteria to the study data, the mean values fall progressively below the optimal level of 100%, and the distributions by medication subgroup become more dissimilar. Table I summarizes the mean values for each treatment group by four definitions; with small sample sizes, none of the intergroup differences is significant at the p < 0.05 level, although all MEMS-based rates are significantly lower than those by pill count. Fig. 2 displays the distribution of interdose intervals by monitor-generated data. Fewer than half of all openings occurred at the prescribed interval of 12 -± 2 hours. Even if one extends the acceptable range to 6 to 18 hours, about 25% of openings took place outside
this generous range.
Impact of scheduled visits. The approach of the scheduled visit occasionally prompted unrepresentative
680 Rudd et
CLIN PHARMACOL THER DECEMBER 1990
al.
DOSES: Last 3 Days'/Whole Interval's (%)
120 100
80 60
113
observations r p
g
0.731 0.001
40 20 0 0
20
40
DAYS:
60
80
100
120
Last 3/Whole Interval (%)
Fig. 3. Homogeneity of drug taking: impact of scheduled visits. Correlation between doses taken (vial openings) and days passed. If scheduled visits stimulate atypical medication-taking behavior, one would expect that the percentage formed by the last 3 days compared with the entire interval between visits would not correlate with the percentage of doses taken in the last 3 days compared with those taken during the entire intervisit interval. Monitor data: n = 18, four to eight visits each.
Table I. Comparison of compliance measures Compliance measure Pill counts (all visits) (%) MEMS monitor (%) Total openings Two openings! day Openings at 10-14-hr intervals
Enalapril
lsradipine
All subjects
1.6*
98.7 ± 2.31-
95.3
2.1t
84.8 6.0t 57.4 ± 6.4t 34.4 ± 4.4t
95.6 ± 2.9§ 69.1 ± 4.9§ 43.1 ± 4.5§
89.5 ± 62.6 ± 38.3 ±
4.411 3.311
91.5
3.8M
Data are mean values -± SEM. Intergroup differences: none significant at 0.05 level. Intragroup differences: all significant at p 0.05. *n = 11; in = 10; in = 21; In = 8; in = 18.
behavior. For some subjects, atypical if near-perfect compliance or apparent overdispensing might occur to make up for previous deficiencies. If so, the traditional pill count would miss the discrepancies, because over dispensing could compensate for previous underdispensing. Fig. 3 exhibits the distribution of matching days with doses. If concordance were perfect, one would expect the number of days between scheduled visits to be proportional to the number of vial openings during the interval. In search of steady-state conditions, we
concentrated on the 3 days before each scheduled return visit, roughly corresponding to five times the half-life of the shorter-acting medication. We then plotted the percentage formed by the 3 days before the scheduled visit to the entire interval in days versus the percentage of openings for the same ratio of times. For example, on the abscissa, 3 days would equal 20% of a 15-day intervisit interval. During the same 3 days, we should expect that 20% of the doses should be dispensed and consumed.
VOLUME 48 NUMBER 6
Monitoring medication compliance TIME OF DAY (24 hr 0100k)
CASE
681
#406
41111.1.11.11 vertigo
2000 1600
1200 -
800
400
3/20
I
3/24
'
i
I
3/28
I
4/1
DATE
I
4/5 VISIT
Fig. 4. Patient-reported vertigo after "hypercompliance" with enalapril with monitor-generated data. Data from medication taking of subject 406, who reported 1 week of vertigo at his return visit on April 7, 1988. Each dot represents one medication vial opening corresponding to one 2.5 mg dosing. On the day before the onset of symptoms, the monitor showed seven openings rather than the usual one to three per day. The symptom, in turn, prompted less medication taking.
Overall, the correlation is high (r = 0.731; p
