Noncompliance in the use of prescribed medication is a well-known problem that may be particularly frequent among pregnant women owing to their fear of fetotoxic side effects of drugs. 1–4 Noncompliance may bias studies in which the exposure information is based solely on prescription data. Misclassification of exposure due to noncompliance may imply that effect estimates predict zero effect levels because the actual exposure is far below the estimated exposure based on purchased prescriptions. 5 Bias toward large effect is seen if prescriptions correlate with use of over-the-counter (OTC) drugs that have the side effects under study.
Reliable data on drug intake are difficult to obtain. 6–9 Differential recall is a serious threat when data are collected after the outcome disease is known by the respondents. 10–13 Prescription data based on registries bypass this recall problem but may be misleading if the prescribed drugs are not used by the patient or are used at a time period different from the date of dispensing. Timing is of essential importance, for example, when studying adverse outcomes of pregnancy, because congenital malformations have a short exposure window of etiologic relevance. Despite these shortcomings, administrative prescription registries are widely used in surveillance of drug safety in pregnancy. 14–24 The main advantages of studies drawing on these databases are the completeness of registration and data that are collected prospectively.
The optimal way to study compliance would be to observe the patients around the clock, which for obvious reasons is impossible. An alternative, although not unproblematic, way of obtaining information about drug intake is to perform interviews about drug use. At best, questions should be asked close to the date of dispensing, and these interviews have to be done by interviewers without any connection to the staff responsible for treatment and without focusing attention on measuring compliance.
We analyzed the extent to which the dispensed medications were reported to be used during pregnancy by the women who were unaware of the intent of this validation exercise.
Subjects and Methods
The study base consisted of all women who were enrolled in the Danish National Birth Cohort (DNBC) in the County of North Jutland from June 1997 through December 1998. All prescriptions purchased by these women were identified in the North Jutland Prescription Database (NJPD). For each woman in the study, we linked data from NJPD with information on drug use provided by the woman to the DNBC.
The personal identification number, which is assigned to all Danish citizens at birth or immigration, enabled us to link information on drug use provided by the women with individual prescription data from the NJPD.
The NJPD includes population-based information on dispensed and refunded prescriptions for the 490,000 inhabitants in the County of North Jutland. 25 All prescribed drugs are sold through monopolized pharmacies equipped with computerized accounting systems linked to the Danish National Health Service. Information on the type of drug prescribed according to the Anatomical Therapeutical Chemical (ATC) classification system, 26 the date of dispensing, and the customer’s personal identification number are electronically transferred to the NJPD from the pharmacies when a prescription is refunded from the Danish National Health Service. Drugs used during hospital admission, nonreimbursed drugs (for example, oral contraceptives, cephalosporines, and tetracyclines), and most of the OTC drugs [such as paracetamol (acetomenaphen), aspirin, low-dose ibuprofen, antihistamines, and certain drugs for topical treatment] are not included in the database. Many prescribed drugs are also available as OTC drugs. In Table 1, drug groups including drugs dispensed both on prescription and as OTC are marked with an asterisk.
The DNBC study aims at recruiting pregnant women as soon as possible after conception to study, among other things, the side effects of medications. The study started as a pilot study in 1994, and the County of North Jutland was enrolled in 1997. The pregnant women are informed about the study by their general practitioner (GP) at the first pregnancy consultation, usually at gestational week 6–12. Together with the informed consent procedure, the pregnant women fill out a questionnaire about drug use for the past 3 months. The questionnaire thus covers periconceptional and early gestational use of medicines. In gestational week 12–15, the pregnant women are interviewed by telephone and asked in detail about their use of medications during pregnancy up to the time of the interview. The telephone interview consists of detailed and structured questions regarding drug use, diseases, and symptoms requiring treatment during pregnancy; the questionnaire data are used as a memory aid for this interview. Responses are coded in fixed answer categories for the most often used drugs and as text strings for the rest. In the pilot study, at least half of the pregnant women received an invitation by their GP, with an overall recruitment of 40% of pregnant women in the County of North Jutland.
Drug Exposure According to the North Jutland Prescription Database and the Danish National Birth Cohort
In the NJPD we identified all prescription drugs dispensed for women in the DNBC within 120 days before the date of interview. NJPDexp gives the number of women who purchased at least one prescription drug. To estimate the impact of recall error in relation to the time to interview, we performed analyses restricted to drugs dispensed 90, 60, and 30 days before the interview.
In the DNBC we defined exposure as reported drug use measured either as yes or no. If a woman used drugs, she was asked to give the brand name, or the type of drug used, or even just the kind of treatment, for example, “some cream” or “a vaginal tablet.” ATC coding of exposure in the DNBC was straightforward when the women gave the brand name of drugs used. More commonly, however, they could only remember the type of treatment as “analgesic,” “cream for treating vaginitis,” or “antibiotic,” but not the drug name. Drugs in DNBC were therefore coded at the third-digit ATC level. DNBCexp gives the number of women who reported use of drugs within ATC groups in the questionnaire and/or in the telephone interview.
The “truth” regarding drug use is not available either in the prescription database or in reported information on drug use. Although repeated measures of drug use are available in the DNBC, there are limitations to the information that are addressed in the discussion. We decided to use the information on drug use in the DNBC as the “referent,” because it aims at addressing actual drug use rather than just the purchase of drugs.
We defined compliance as the agreement between registrations in the NJPD and the DNBC, measured as the predictive value of a positive registration in NJPD. Thus, compliance in this study describes the probability of reporting drug use in DNBC after a dispensed prescription. We analyzed compliance on the basis of drugs dispensed 120, 90, 60, and 30 days before the interview.
In the County of North Jutland 2,041 women were interviewed in the DNBC study from June 1997 through December 1998. In the NJPD we identified 1,623 prescription drugs purchased by 796 of the 2,041 women in the DNBC. Thus, 39% of the cohort participants purchased at least one prescription drug during the 120 days before the interview.
The overall compliance for drugs purchased within 120 days before the interview was 43% [95% confidence interval (CI) = 40–46]. A narrowing of the time period revealed compliance of 47% (95% CI = 43–50) 90 days before the interview, 49% (95% CI = 44–53) 60 days before the interview, and 50% (95% CI = 43–56) 30 days before the interview. Compliance was high (70–100%) for drugs used in the treatment of chronic diseases such as insulin-dependent diabetes mellitus, thyroid diseases, epilepsy, depression, asthma, and hypertension (diuretics and beta-blocking agents), and for prescribed analgesics. Compliance was lower for short-term treatment, for example, with antihistamines, antibiotics, antacids, anti-infective drugs for gynecologic use, and nonsteroid anti-inflammatory drugs. The lowest compliance was seen for local treatments of cough and cold and for use of ophthalmologic and dermatologic preparations.
An average of 43% of all drugs dispensed according to the prescription database were reported to be used, but this value covers a range from 0% for certain local treatments to 100% for drugs used for chronic diseases.
No single source of information on drug use is both complete and specific. The NJPD has complete and valid data on dispensed drugs, but it is only a surrogate measure for drug use. For example, the NJPD does not include OTC drug, drugs not reimbursed, and drugs used during hospital admission. Moreover, we have only marginal knowledge of the timing of drug use, as we only have access to the date of dispensing. The DNBC aims at covering all drug intake. Data on drug use in the DNBC are based partly on the questionnaire, which may be less valid than information obtained by interview. 6,7,27,28 We know that self-reports on drug use suffer from imperfect recall and even denial.
In the DNBC study, the women were first asked open-ended questions; thus, they could report drug use with their own words. Furthermore, the questions were specific, such as “Did you receive any treatment for vaginitis?” It is difficult to imagine a study with better data on actual drug use without interfering too much with behavior. Drug use during pregnancy is a subject of concern, and we do find it reasonable to believe that the pregnant women remembered what they actually used (also, prescription steroid lotions and local treatments for vaginitis). Nevertheless, we expect the DNBC data to suffer from lack of recall or unwillingness to report drug use to the interviewers, 9 which will cause compliance to be underestimated, especially for short-term or local treatments. On the other hand, the time period of recall was short, and restricting analyses to prescriptions dispensed 60 or even 30 days before the interview only slightly improved compliance. It is therefore unlikely that the discrepancies found are entirely due to lack of recall.
Results may differ from what we would see among pregnant women at large, given that only about 40% of the pregnant women were enrolled into the DNBC study. Women were invited to participate in the DNBC study by the GP at the first pregnancy consultation, and we expect about half of the GPs to take part in the study. If a woman decided to participate, she filled out and returned the questionnaire together with the informed consent form. We expect participating women to be more “compliant” than nonparticipating women. The percentage of women to whom prescription drugs were dispensed and the type of drugs dispensed in the present study are comparable with results from a previous study concerning dispensed prescriptions among pregnant women in the county of North Jutland. 29
We may also have overestimated compliance by the way we analyzed the data, given that reported use of OTC drugs may be categorized in the same ATC group as a purchased prescription drug that was not used. Prescriptions were classified within time windows according to the date of the interview. In the questionnaire, the women were asked to report use of drugs 3 months before and during early pregnancy. If the time difference between the questionnaire and the interview exceeded 120 days, the information from the questionnaire concerned prescription drugs that were dispensed before the study time window. We did not have information on the date of filling out the questionnaire for all of the women in the study base, but most interviews were performed in weeks 12–15 of gestation. We therefore think it unlikely that differences in time span between questionnaire and interview should bias the results; if so, the bias will tend to overestimate compliance.
Recent literature reviews have described compliance ranging from 7% to 85%. 1–3 The disparity between the estimates can be attributed to the population studied, the medical regimen, and definitions of compliance and methodology. We found that compliance was high for drugs used for certain chronic diseases. This finding agrees with previous studies of compliance among patients with epilepsy and diabetes mellitus. 4,30 Our finding of compliance for asthma drugs was higher than reported in previous studies. 31–34
In conclusion, we found that overall 43% of the dispensed prescription drugs were reported as actually used by the pregnant women. Exposure misclassification due to noncompliance may produce false negative results in risk-assessment studies in which exposure is based solely on prescription, dispensing, or health insurance data. Our study indicates that for some drug groups the computer-recorded prescription may be a reliable measure for exposure, but for others it is not. Compliance to the drug being evaluated should therefore be considered individually.
EUROMAP comprises the following members. Project management group: Henrik Toft Sørensen, Jørn Olsen (Project Leader), Andrew Czeicel, Gunnar Lauge Nielsen, Lolkje De Jong-van den Berg, Lorentz Irgens, Ulf Bergman, and Charlotte Olesen (Project Manager). Other members: Lars Pedersen, Rolv T. Lie, Corinne de Vries, Helle Larsen, and Jørgen Bendtsen.
We thank the staff at the Department of Health Insurance and Preventive Medicine and Hospital Discharge Registries in the County of Northern Jutland and at the Danish Epidemiology Science Centre, where the National Birth Cohort was organized, for excellent assistance in preparing the data for analyses.
1. The magnitude of adherence. In: Haynes TB, Taylor DW, Sackett DL, eds. Compliance
in Health Care. Baltimore: John Hopkins University Press, 1979;11–22.
2. Greenberg RN. Overview of patient compliance
with medication dosing: a literature review. Clin Ther 1984; 6: 592–599.
3. Christensen DB. Drug-taking compliance
: a review and synthesis. Health Serv Res 1978; 13: 171–187.
4. Cramer JA, Mattson RH, Prevey ML, Scheyer RD, Ouellette VL. How often is medication taken as prescribed? A novel assessment technique. JAMA 1989; 261: 3273–3277.
5. Graham DJ, Smith CR. Misclassification in epidemiologic studies of adverse drug reactions using large managerial data bases. Am J Prev Med 1988; 4 (2 suppl):15–24.
6. de Jong van den Berg LT, Waardenburg CM, Haaijer Ruskamp FM, Dukes MN, Wesseling H. Drug use in pregnancy
: a comparative appraisal of data collecting methods. Eur J Clin Pharmacol 1993; 45: 9–14.
7. de Jong van den Berg LT, Feenstra N, Sørensen HT, Cornel MC. Improvement of drug exposure data in a registration of congenital anomalies. Pilot-study: pharmacist and mother as sources for drug exposure data during pregnancy
. EuroMAP Group. European Medicine and Pregnancy
Group. Teratology 1999; 60: 33–36.
8. Hertz Picciotto I. Agreement between questionnaire data and medical records: the evidence for accuracy of recall. Am J Epidemiol 1991; 133: 408–410.
9. Mitchell AA, Cottler LB, Shapiro S. Effect of questionnaire design on recall of drug exposure in pregnancy
. Am J Epidemiol 1986; 123: 670–676.
10. Bar-Oz A, Moretti ME, Mareels G, Tittelboom TV, Koren G. Reporting bias in retrospective ascertainment of drug-induced embryopathy. Lancet 1999; 354: 1700–1701.
11. Swan SH, Shaw GM, Schulman J. Reporting and selection bias in case-control studies of congenital malformations. Epidemiology 1992; 3: 356–363.
12. Bryant HE, Visser N, Love EJ. Records, recall loss, and recall bias in pregnancy
: a comparison of interview and medical records data of pregnant and postnatal women. Am J Public Health 1989; 79: 78–80.
13. Feldman Y, Koren G, Mattice K, Shear H, Pellegrini E, MacLeod SM. Determinants of recall and recall bias in studying drug and chemical exposure in pregnancy
. Teratology 1989; 40: 37–45.
14. Ruigómez A, García Rodríguez LA, Cattaruzzi C, Troncon MG, Agostinis L, Wallander MA, Johansson S. Use of cimetidine, omeprazole, and ranitidine in pregnant women and pregnancy
outcomes. Am J Epidemiol 1999; 150: 476–481.
15. Briggs GG, Freeman RK, Summer JY. In: Mitchell CW, ed. Drugs in Pregnancy
and Lactation. Baltimore: Williams & Wilkins, 1994.
16. Piper JM, Baum C, Kennedy DL. Prescription drug use before and during pregnancy
in a Medicaid population. Am J Obstet Gynecol 1987; 157: 148–156.
17. Brocklebank JC, Ray WA, Federspiel CF, Schaffner W. Drug prescribing during pregnancy
. A controlled study of Tennessee Medicaid recipients. Am J Obstet Gynecol 1978; 132: 235–244.
18. Wilton LV, Pearce GL, Martin RM, Mackay FJ, Mann RD. The outcomes of pregnancy
in women exposed to newly marketed drugs in general practice in England. Br J Obstet Gynaecol 1998; 105: 882–889.
19. Correy JF, Newman NM, Collins JA, Burrows EA, Burrows RF, Curran JT. Use of prescription drugs in the first trimester and congenital malformations. Aust NZ J Obstet Gynaecol 1991; 31: 340–344.
20. Olesen C. Prescribing during pregnancy
and lactation with reference to the Swedish classification system. A population-based study among Danish women. Eur J Obstet Gynecol 1999; 78: 1–7.
21. Sorensen HT, Steffensen FH, Olesen C, Nielsen GL, Pedersen L, Olsen J. Pregnancy
outcome in women exposed to calcium channel blockers. Reprod Toxicol 1998; 12: 383–384.
22. Steffensen FH, Nielsen GL, Sorensen HT, Olesen C, Olsen J. Pregnancy
outcome with ACE-inhibitor use in early pregnancy
. Lancet 1998; 351: 596–596.
23. Sorensen HT, Nielsen GL, Olesen C, Larsen H, Steffensen FH, Schonheyder HC, Olsen J, Czeizel AE. Risk of malformations and other outcomes in children exposed to fluconazole in utero
. Br J Clin Pharmacol 1999; 48: 234–238.
24. Nielsen GL, Sorensen HT, Thulstrup AM, Tage-Jensen U, Olesen C, Ekbom A. The safety of proton pump inhibitors in pregnancy
. Aliment Pharmacol Ther 1999; 13: 1085–1089.
25. Olesen C, Steffensen FH, Nielsen GL, de Jong van den Berg LT, Olsen J, Sørensen HT. Drug use in first pregnancy
and lactation. A population-based survey among Danish women. Eur J Clin Pharmacol 1999; 55: 145–149.
26. Introduction. Drug Utilization Studies: Methods and Uses. World Health Organization Regional Publications, European Series, No. 1993, Pub No. 45. Geneva: World Health Organization, 1993.
27. de Jong PC, Huijsmans AA, Nienhuis HE, Nijdam WS, Zielhuis GA, Eskes TK. Validation of a questionnaire on medical drug use during pregnancy
. Am J Epidemiol 1991; 134: 998–1002.
28. Harlow SD, Linet MS. Agreement between questionnaire data and medical records. The evidence for accuracy of recall. Am J Epidemiol 1989; 129: 233–248.
29. Olesen C, Steffensen FH, Nielsen GL, de Jong-van den Berg L, Olsen J, Sørensen HT. Drug use in first pregnancy
and lactation: A population-based survey among Danish women. Eur J Clin Pharmacol 1999; 55: 139–144.
30. Takaki S, Kurokawa T, Aoyama T. Monitoring drug noncompliance in epileptic patients: assessing phenobarbital plasma levels. Ther Drug Monit 1985; 7: 87–91.
31. Yeung M, O’Connor SA, Parry DT, Cochrane GM. Compliance
with prescribed drug therapy in asthma. Respir Med 1994; 88: 31–35.
32. Spector SL, Mawhinney H. More on patient compliance
. J Allergy Clin Immunol 1989; 84: 409–410.
33. Creer TL, Levstek D. Medication compliance
and asthma: overlooking the trees because of the forest. J Asthma 1996; 33: 203–211.
34. Bosley CM, Fosbury JA, Cochrane GM. The psychological factors associated with poor compliance
with treatment in asthma. Eur Respir J 1995; 8: 899–904.