QUICK SEARCH:   [advanced] Author: Keyword(s): Year:  Vol:  Page:

This Article Abstract Full Text (PDF) Purchase Article View Shopping Cart Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Jasti, S. Articles by Hartzema, A. G. Search for Related Content PubMed PubMed Citation Articles by Jasti, S. Articles by Hartzema, A. G.

---

Methodology and Mathematical Modeling

Correction for Errors in Measuring Adherence to Prenatal Multivitamin/Mineral Supplement Use among Low-Income Women^1,2

Sunitha Jasti^*,3, Anna Maria Siega-Riz, Mary E. Cogswell^** and Abraham G. Hartzema

^* Department of Family, Nutrition and Exercise Sciences, Queens College of City University of New York, Flushing, NY 11367; Department of Nutrition, School of Public Health, the University of North Carolina at Chapel Hill and Carolina Population Center, Chapel Hill, NC 27516; ^** Maternal and Child Nutrition Branch, Division of Nutrition and Physical Activity, Centers for Disease Control and Prevention, Atlanta, GA 30341; and Pharmacy Health Care Administration, University of Florida, Gainesville, FL 32610

³ To whom correspondence should be addressed. Email: sunitha_jasti{at}qc.edu.

	ABSTRACT

TOP ABSTRACT SUBJECTS AND METHODS RESULTS DISCUSSION LITERATURE CITED

 
Adherence to prenatal multivitamin/mineral supplement use is often measured by self-reports or pill counts. Although both measures were shown to overestimate adherence, measurement error is rarely considered. In this study, we examined measurement error in adherence to prenatal supplement use among pregnant women and demonstrated a calibration method to adjust for error. In a validation subsample (n = 51) from a larger clinical study of supplementation, adherence was assessed by self-reports, pill counts, and a Medication Event Monitoring System (MEMS) bottle cap that recorded the date and time of each opening of the pill bottle. Mean adherence in the validation sample as measured by the MEMS (the gold standard) was 68%; thus, adherence measured by self-report (77%) and pill count (84%) reflected overestimation. The Pearson correlation coefficients of self-reports and pill counts to MEMS were 0.35 and 0.62, respectively. When adherence was defined as taking 75% of the pills prescribed, sensitivity and specificity were greater for pill counts (93 and 52%, respectively) than for self-reports (88 and 44%). The regression coefficient for pill count adherence from a linear regression on MEMS adherence was applied to pill counts from a larger sample (n = 244). The adjustment significantly lowered the estimate of adherence from 74 to 64% (P < 0.001) in this larger sample. In conclusion, our data show that both self-reports and pill counts overestimate adherence and that linear regression in comparison to an external standard such as MEMS can be used to correct for measurement error in adherence.

KEY WORDS: • adherence • measurement error • prenatal supplements

In the United States, multivitamin/mineral supplements containing iron are prescribed routinely for pregnant women to prevent anemia (1). The evidence for beneficial effects of iron supplementation during pregnancy on outcomes such as preterm birth or low birth weight, however, has been equivocal (2–4). Low adherence to iron supplement use was suspected to have produced false negative results in some of the studies (4). Analysis by intention-to-treat remains the standard methodology in randomized clinical trials (RCT)⁴ with the assumption that the likelihood of nonadherence to treatment or loss to follow-up among the subjects also is randomly distributed across treatment groups. In some studies, measures are taken to maintain high adherence levels, but adherence to treatments is rarely considered in analysis. Supplementary analyses that incorporate adherence data in RCT are usually considered difficult to interpret due to the potential selection bias and the measurement error inherent in adherence data (5,6). However, increasingly more sophisticated statistical approaches to model the effect of adherence in RCT are reducing the biases involved and are helping to provide additional insights into true treatment effects (7–10). Nevertheless, accurate assessment of adherence is crucial and relevant in understanding the true effectiveness of supplements such as iron during pregnancy. Despite the importance of adherence in clinical care and RCT, research on adherence measurement techniques is limited, particularly in studies involving treatment or prevention of certain conditions during pregnancy.

The majority of information on iron supplement use in the United States is based on self-reported data and was collected using broad questions (11–13). For example, women in the National Maternal and Infant Health Survey (12) were considered users if they reported taking multivitamin/mineral supplements at least 3 d a week during the 3 mo after they found out about their pregnancy. In the National Health and Nutrition Examination Survey III, women reporting use of any supplements containing iron in the month before the interview time were considered to be users (13). Responses to such questions potentially overestimate supplement use and adherence, especially when participants are asked to remember their use of supplements over long periods of time (14). Self-reported adherence to iron supplements was overestimated when compared with stool tests for iron among pregnant women in developing countries (15,16), but the measurement error was rarely considered and accounted for in analysis. Despite the problems associated with self-reports such as social desirability bias and forgetfulness, they may be the only practical way to assess adherence because self-reports are easier and cheaper to obtain than other measures such as pill counts, refill rates from pharmacy records, electronic monitors, and biomarkers. Hence, ways to improve estimation of adherence using self-reported data are of interest.

In the medication adherence literature, comparisons have been made among self-reports, pill counts, the Medication Event Monitoring System (MEMS), an electronic monitoring device, viral loads, and innovative approaches to improve measurement of adherence to HIV antiretroviral treatments (17–19). Self-report adherence was higher than the MEMS adherence (78 vs 53%, respectively) to antiretroviral therapy in current and former drug users, although both measures strongly correlated with viral loads (17). A composite adherence score constructed by combining MEMS, pill count, and interview data was found to have the strongest predictive relation with achievement of undetectable viral load within 6 mo of initiating therapy (18). Detailed comparisons of different measures of adherence to iron supplements and quantitative descriptions of the measurement error involved are limited in the United States. In this study, we examined measurement error in adherence to the use of prenatal multivitamin/mineral supplements containing iron among low-income pregnant women and demonstrated a calibration method to adjust for the error.

	SUBJECTS AND METHODS

TOP ABSTRACT SUBJECTS AND METHODS RESULTS DISCUSSION LITERATURE CITED

 
    Study population and data collection. Data used in this study were from the iron supplementation study, a randomized, double-blind, clinical control study that was designed to test the effectiveness of the Institute of Medicine recommendations for the prevention of iron deficiency in the 3rd trimester of pregnancy (20). The study recruited women (n = 967, 60% participation rate) receiving care at the Wake County Human Resources prenatal clinic, in Raleigh, NC, between 1997 and 1999. Baseline information on sociodemographic characteristics and pregnancy history was abstracted from medical records. Additional information on prenatal, intrapartum, and birth outcome information was abstracted after delivery. The study design, eligibility criteria, and losses to follow-up were described in detail elsewhere (21). The study was approved by the Institutional Review Board of The University of North Carolina, School of Medicine; the Institutional Review Committee of Wake Medical Center; and the Institutional Review Board for Human Subjects at the CDC.

    Assessment of adherence. Self-report adherence was calculated from responses to a series of questions about usage during the past week. Because the questionnaire did not cover taking pills on d 1, 2, and 3 of the past week, a continuous variable could not be constructed for adherence during the entire week. Adherence over 4 d (today plus the previous 3 d) was constructed as a continuous variable, by considering only the responses to questions on the 4 d of interest and using the formula: adherence = [(number of pills taken in 4 d)/4] x 100. A categorical variable was constructed reflecting 3 levels of adherence during 1 wk, i.e., no adherence (no pills taken), some adherence, and full adherence (all pills taken).

To assess adherence by pill counts, study participants were given bottles containing 32 pills and instructed to return the bottle with unused pills at their next prenatal visit. Pill count adherence was calculated using the following formula:

Pill count adherence = [(32 – number of pills left in the bottle) x 100]/(number of days between dispensing date and return date of pill bottle).

For the validation study, a subsample of women (100 consecutively randomized women during a randomly picked period of recruitment) received 90 pills in MEMS bottles and were instructed to return the bottle with unused pills either at their 26–28 wk visit or when they ran out of pills, whichever occurred first. The formula for pill count adherence for this subsample (the primary study group for the present report) was therefore modified as follows:

Pill count adherence = [(90 – number of pills left in the bottle) x 100]/(number of days between dispensing date and return date of pill bottle).

As the gold standard, adherence was calculated using the data from the MEMS caps, which have an electronic counting device that records the date and time of every opening. Here, adherence was defined as the percentage of pills taken as prescribed in the analysis period, and this rate was used to test the validity of self-report and pill count measures. The number of openings recorded by MEMS was assumed to correspond to the number of pills consumed.

Adherence per the MEMS was calculated for the validation sample only and was for 2 different periods of analysis: 1) the entire period that women used the bottle (MEMS90), and 2) the week before the bottle was returned (MEMS7):

1) MEMS90 = [(90 – number of openings) x 100]/(number of days between dispensing date and return date of pill bottle.

2) MEMS7 = [(7 – number of openings during the week before bottle return) x 100]/7.

Women in the validation sample also provided self-report and pill count measures of adherence as noted. Of the 967 women who were recruited, only 549 (57%) actually participated in the study because they were the only ones who picked up the prenatal supplements provided by the study at the clinic pharmacy. Of these 549, a total of 437 returned for a refill and provided a self-reported measure of adherence by completing a questionnaire. Of the 100 women who participated in the validation study, only 59 returned the bottles after use. Of 437 women, 252 returned their regular pill bottles, but only 244 had pill count adherence calculated, and 51 of these were MEMS women. Thus, in this study, the sample sizes were as follows: 51, validation sample; 244, pill count; and 437, self-report.

    Statistical analysis. A comparison of characteristics of the women in the validation, self-report (minus those who were also in the validation group), and pill count samples (again minus the validation group) was performed by ² tests for categorical variables and t tests for continuous variables. Univariate statistics for each adherence measure were calculated and correlations between different adherence measures were examined. Agreement between MEMS, pill count, and self-report adherence measures was examined by defining adherence as at least 75% of pills taken as prescribed. Sensitivity, specificity, and positive predictive value were calculated for pill count and self-report measures using MEMS as the gold standard. Simple linear regression was used to determine the bivariate association between MEMS90 and pill counts. Multivariate linear regression was employed to calibrate the pill count measure to MEMS adherence. Maternal characteristics, including age, ethnicity, marital status, gravidity, education, and gestational age at entry to prenatal care and at adherence assessment were tested for confounding and controlled for when confounding was significant. A simple linear approximation method was used to obtain calibrated pill count, with the assumption that the error in pill count was not correlated with that in the MEMS measures. Regression diagnostics were performed to determine the stability of the models. Because the distribution of each of our adherence variables was skewed to the right, we examined plots of residuals against fitted values and found that variance was constant for the final model. Calibration using a linear approximation method could not be performed on self-report data due to the small range of the continuous variable for self-report adherence. All analyses were performed using statistical software package STATA version 8.

	RESULTS

TOP ABSTRACT SUBJECTS AND METHODS RESULTS DISCUSSION LITERATURE CITED

 
    Sample characteristics. Of the 51 women in the validation sample, 57% were non-Hispanic Black, and 33% were non-Hispanic White (Table 1). Maternal age ranged from 13 to 37 y (mean: 23 y). For 39% of women, it was their first pregnancy, and most (88%) were single. Only 26% had >12 y of education. Mean gestational age at entry to prenatal care was 13.1 wk ranging from 8 to 19 wk (data not shown).

    Measurement error in pill count and self-report adherence. Within the validation sample, adherence per MEMS90 (68%) was lower than that obtained from the pill count (84%) (P < 0.001) and self-report (77%) (Table 2), but the difference from MEMS was not significant for self-report (P = 0.05). The shorter MEMS measure (MEMS7) produced a slightly lower estimate (by 4 percentage points) than MEMS90, and was lower than the adherence measured by self-report (P < 0.02). The means for pill count (83.9%) and self-report (76.5%) did not differ (P = 0.11). When comparisons were made (within the validation sample) among groups on the proportion with full (100%) adherence, however, a larger proportion of women were classified as having full adherence by the self-report measure (63%) than by pill count (20%) (P < 0.01) or MEMS90 (0%) (Table 2), which is consistent with the much shorter analytic period for self-report.

View this table: [in this window] [in a new window]   TABLE 4 Agreement between MEMS, pill count, and self-report measures in low-income women receiving prenatal vitamin/mineral supplementation when adherence was defined as 75% pills taken as prescribed1

	DISCUSSION

TOP ABSTRACT SUBJECTS AND METHODS RESULTS DISCUSSION LITERATURE CITED

The correlation between self-report and pill count adherence found in our study (0.41) is much lower than that reported in an Australian study (0.86) of 338 pregnant women prescribed iron tablets (22). In our study, the difference in time periods between self-report adherence (4 d) and pill count adherence (90 d) may have decreased the correlation. We examined the correlation of self-report adherence with the MEMS adherence measured during 1 wk before bottle return (MEMS7) to reduce this disparity in assessment duration. The correlation (between self-report adherence and MEMS7) was not substantially better (0.46). This suggests that in our study, the difference in time periods was not a factor in the low correlation observed between self-reported adherence and MEMS measured adherence. The Australian study also reported a much higher pill count adherence than that found in our study sample, i.e., >85% of the Australian study participants took all of the pills as prescribed (100% adherence), compared with 20% in our validation sample. Monthly telephone calls made to encourage adherence in the Australian study may have led to the high adherence levels. It is difficult to interpret the difference, however, because few details were given about the adherence measures.

The different results from comparison of self-report and pill count measures in terms of mean adherence (pill counts higher than self-reports) and adherence in categories (self-reports higher than pill counts) in our study send a cautionary message. Adherence data collected in a continuous form over a longer period, which can then be categorized for comparisons, are more appropriate and useful than data collected in categories, which is the norm in supplement use studies (11,12). When adherence was defined in broad categories of 75% and < 75% of supplements taken as prescribed, we found that the pill count measure still had a higher sensitivity, specificity, and a positive predictive value than the self-report measure.

Because self-reports are most commonly used to assess adherence and most often are the only feasible approach in community supplementation trials, application of the calibration method we described to adjust for measurement error in pill counts, to self-report data in a continuous form, may offer a way to improve adherence data in studies with limited time and financial resources. This method was used by Liu et al. (18) in developing a composite adherence score that uses MEMS data as its backbone while imputing pill count or self-report data when MEMS measure is missing to measure adherence to HIV treatment. They adjusted for overestimation by self-report and pill count by calibrating them to MEMS adherence before incorporating them into the composite score. The composite score predicted undetectable viral load better than each of the adherence measures used in its calculation (18). Self-report had unstable estimates of the effect and was not significantly related to achieving viral suppression (18). In our study, we were unable to regress self-report adherence on MEMS adherence due to the limited range in the self-report adherence (4 d). Thus, collection of continuous data on adherence over at least a 1-wk period may better capture the range in adherence levels, especially if assessed at regular intervals during the entire pregnancy and may be more favorable for calibration purpose.

We attempted to assess adherence by both self-reports and pill counts at 4-wk intervals in our study, which coincided with the intervals between prenatal clinic appointments of our participants. However, due to the high rate of loss to follow-up, we were able to use only data from the first adherence assessment, which was available for only 50% of the women who took part in the study. Even though analysis of data on multiple pill counts in those women who did keep most of their clinic appointments showed that within-person variability was low (data not shown) (21), it is possible that self-report adherence over 1 wk does not accurately represent the adherence over the entire supplementation period.

Losses in our study sample may have induced selection bias as implied by subgroup comparisons. For example, self-report adherence was higher when pill counts were provided than when they were absent (Table 1) in the larger study sample (n = 437), whereas pill count adherence was higher in the validation sample than in the larger sample (Table 1), suggesting that adherence measures in our study may be higher than the true adherence levels in the clinic population.

Because the MEMS caps could not be reused once data were read, and we wanted to assess and compare adherence over a longer period of time, MEMS bottles were dispensed with 90 pills rather than the 32 pills in the regular bottles. However, because we were unable in the end to use multiple pill counts and self-reports for comparison with MEMS adherence in each person due to high loss to follow-up, a large difference in analysis period was created between self-report and pill count measures. Using MEMS as the more objective gold standard has its limitations as well. Each opening of the bottle is assumed to be a dose taken. Participants were not given an explanation of specific function of the MEMS cap, but were offered $5 as an incentive for returning the bottle. If they unwittingly decanted multiple pills at one time for doses to be taken on the subsequent days (not wanting to carry the large pill bottle on their short trips out of town), MEMS underestimated their adherence.

Despite the limitations in our study, the adjustment for measurement error by calibration to a more objective adherence measure (as we demonstrated in our study using MEMS) may reduce the overestimation by self-reported data used in most trials of supplementation. Further research on ways to reduce selection bias from the influence of MEMS caps on adherence and longitudinal data on adherence may improve the validity of calibrated measures of adherence.

	FOOTNOTES

 
¹ Supported by a Cooperative Agreement with the Association of Schools in Public Health/Centers for Disease Control and Prevention (#S454).

² The findings and conclusions in this report are those of the author(s) and do not necessarily represent the views of the funding agency.

⁴ Abbreviations used: MEMS, Medication Event Monitoring System; MEMS7, adherence measured by MEMS during the week before bottle return; MEMS90, adherence measured by MEMS during the entire 90-d period the bottle was used, RCT, randomized controlled trial.

Manuscript received 10 August 2005. Initial review completed 27 September 2005. Revision accepted 2 November 2005.

	LITERATURE CITED

TOP ABSTRACT SUBJECTS AND METHODS RESULTS DISCUSSION LITERATURE CITED

 

1. Centers for Disease Control and Prevention. Recommendations to prevent and control iron deficiency in the United States. MMWR Recomm Rep. 1998 Apr 3;47(RR-3):1–29.[Medline]

2. US Preventive Services Task Force. Routine iron supplementation during pregnancy. Policy statement. JAMA. 1993;270:2846–8.[Medline]

3. Allen LH. Pregnancy and iron deficiency: unresolved issues. Nutr Rev. 1997;55:91–101.[Medline]

4. Rasmussen K. Is There a Causal Relationship between Iron Deficiency or Iron-Deficiency Anemia and Weight at Birth, Length of Gestation and Perinatal Mortality? J Nutr. 2001 Feb;131(2S–2):590S–601; discussion S-3S.[Abstract/Free Full Text]

5. Lee YJ, Ellenberg JH, Hirtz DG, Nelson KB. Analysis of clinical trials by treatment actually received: is it really an option? Stat Med. 1991;10:1595–605.[Medline]

6. Newell DJ. Intention-to-treat analysis: implications for quantitative and qualitative research. Int J Epidemiol. 1992;21:837–41.[Abstract/Free Full Text]

7. Efron B, Feldman D. Compliance as an explanatory variable in clinical trials. J Am Stat Assoc. 1991;86:9–17.

8. Cuzick J, Edwards R, Segnan N. Adjusting for non-compliance and contamination in randomized clinical trials. Stat Med. 1997;16:1017–29.[Medline]

9. Robins JM. Correcting for non-compliance in randomized trials using structural nested means models. Commun Stat Theory Methods. 1994;23:2379–412.

10. Nagelkerke N, Fidler V, Bernsen R, Borgdorff M. Estimating treatment effects in randomized clinical trials in the presence of non-compliance. Stat Med. 2000;19:1849–64.[Medline]

11. Orr RD, Simmons JJ. Nutritional care in pregnancy: the patient's view. II. Perceptions, satisfaction, and response to dietary advice and treatment. J Am Diet Assoc. 1979;75:131–6.[Medline]

12. Yu SM, Keppel KG, Singh GK, Kessel W. Preconceptional and prenatal multivitamin-mineral supplement use in the 1988 National Maternal and Infant Health Survey. Am J Public Health. 1996;86:240–2.[Abstract/Free Full Text]

13. Cogswell ME, Kettel-Khan L, Ramakrishnan U. Iron supplement use among women in the United States: science, policy and practice. J Nutr. 2003;133:1974S–7.[Abstract/Free Full Text]

14. Norell SE. Accuracy of patient interviews and estimates by clinical staff in determining medication compliance. Soc Sci Med [E]. 1981;15:57–61.[Medline]

15. Schultink W, van der Ree M, Matulessi P, Gross R. Low compliance with an iron-supplementation program: a study among pregnant women in Jakarta, Indonesia. Am J Clin Nutr. 1993;57:135–9.[Abstract/Free Full Text]

16. Bondarianzadeh D, Siassi F, Omidvar N, Golestan B, Keighobadi K. Low compliance with iron supplementation program among pregnant women in the rural areas of Kerman district, I.R. Iran. Nutr Res. 1998;18:945–52.

17. Arnsten JH, Demas PA, Farzadegan H, Grant RW, Gourevitch MN, Chang CJ, Buono D, Eckholdt H, Howard AA, Schoenbaum EE. Antiretroviral therapy adherence and viral suppression in HIV-infected drug users: comparison of self-report and electronic monitoring. Clin Infect Dis. 2001;33:1417–23.[Medline]

18. Liu H, Golin CE, Miller LG, Hays RD, Beck CK, Sanandaji S, Christian J, Maldonado T, Duran D, et al. A comparison study of multiple measures of adherence to HIV protease inhibitors. Ann Intern Med. 2001;134:968–77.[Abstract/Free Full Text]

19. Kimmerling M, Wagner G, Ghosh-Dastidar B. Factors associated with accurate self-reported adherence to HIV antiretrovirals. Int J STD AIDS. 2003;14:281–4.[Abstract/Free Full Text]

20. Institute of Medicine, Food and Nutrition Board. Iron deficiency anemia: guidelines for prevention, detection and management among U.S. children and women of childbearing age. Washington (DC: National Academy Press, 1993.

21. Jasti S, Siega-Riz AM, Cogswell ME, Hartzema AG, Bentley ME. Pill count adherence to prenatal multivitamin/mineral supplement use among low-income women. J Nutr. 2005;135:1093–101.[Abstract/Free Full Text]

22. Makrides M, Crowther CA, Gibson RA, Gibson RS, Skeaff CM. Efficacy and tolerability of low-dose iron supplements during pregnancy: a randomized controlled trial. Am J Clin Nutr. 2003;78:145–53.[Abstract/Free Full Text]

This article has been cited by other articles:

				M. A. Chisholm-Burns and C. A. Spivey Pharmacoadherence: A new term for a significant problem Am. J. Health Syst. Pharm., April 1, 2008; 65(7): 661 - 667. [Full Text] [PDF]

				A. L. Quittner, A. C. Modi, K. L. Lemanek, C. E. Ievers-Landis, and M. A. Rapoff Evidence-based Assessment of Adherence to Medical Treatments in Pediatric Psychology J. Pediatr. Psychol., September 10, 2007; (2007) jsm064v1. [Abstract] [Full Text] [PDF]

This Article Abstract Full Text (PDF) Purchase Article View Shopping Cart Alert me when this article is cited Alert me if a correction is posted Services Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Jasti, S. Articles by Hartzema, A. G. Search for Related Content PubMed PubMed Citation Articles by Jasti, S. Articles by Hartzema, A. G.