A comparison of cross-sectional and longitudinal survey methods in measuring patient compliance with tuberculosis treatment.

A group of 179 tuberculosis patients, notified within a specific time period in an area served by a community-based TB clinic in Cape Town, was defined. Cross-sectional and longitudinal measurements of compliance were taken to test the validity of the two research methods in compliance surveillance. The cross-sectional method was found to mask significantly the true extent of non-compliance with tuberculosis treatment (19.5% v. a true 40.3%). The sampling and measurement biases inherent in the cross-sectional design were illustrated. The first bias was introduced because the cross-sectional survey captured only treatment survivors. By the date of the cross-sectional survey 45 patients were no longer on the clinic treatment list, between 4 and 13 (2.2 - 7.3% of the original group) having been lost as a result of poor compliance. The second and most extensive bias was introduced because the cross-sectional survey measured the compliance of all patients on a particular day irrespective of the time since notification. Thus for the cross-sectional group the compliant proportion was 80.5% at the time of the cross-sectional survey but fell to 65.4% by the end of treatment (as measured in the longitudinal survey). The absolute significance of the association between compliance and various demographic and treatment variables did not change with survey method. However, in two instances the association approached significance in only one of the two surveys. The study implies that serious clinical repercussions are possible if management decisions are based on the results of inappropriately designed studies. The place of the longitudinal and cross-sectional research methods in compliance surveillance is discussed.