Investigating the effect of nonparticipation using a population-based case–control study on myocardial infarction

https://doi.org/10.1016/j.annepidem.2003.09.013Get rights and content

Abstract

Purpose

To characterize non-participants in a population-based study on cardiovascular diseases and investigate the effect of non-participation on risk estimates for myocardial infarction.

Methods

Using random digit dialing we obtained full information for 1054 adults (60.8% female), while 345 eligible individuals (72.5% female) declined the invitation to participate, but answered a limited set of questions by telephone. Risk of myocardial infarction was estimated using 474 cases (19.4% females) admitted with a first acute myocardial infarction.

Results

Participation proportion was 99.0% for cases and 70.0% for population controls. Population non-participants were older (61.6 vs. 58.5 years, for males, and 62.9 vs. 57.7 years for females) and more frequently women (66.3% vs. 74.7%, p < 0.001); males tended to be non-drinkers and to have had a blood test during the previous year; females were additionally more often non-smokers. Crude and adjusted risk estimates for myocardial infarction were generally similar regardless of considering the information provided by non-participants.

Conclusions

In this South European population, demographic and social characteristics associated with the decision to participate in a community investigation were different from those usually described in Northern European or American populations. However, their characteristics did not influence the direction or the magnitude of myocardial infarction risk estimates.

Introduction

Recruitment strategies to avoid participation bias are essential 1., 2.. The ideal situation would be to keep refusal as close to zero as possible but in general not every eligible person will participate. Participation proportion is related to design options, study topics, and cultural background of the population (3). As the latter might add specific concerns we cannot assume that determinants of non-participation operate everywhere in the same way.

When refusal to participate is differential there is a particular risk of biased estimates. This is frequently expected when population-based control groups are used or general sampling procedures, such as random digit dialing, is the option 4., 5., 6..

If the decision to participate in an epidemiological investigation is associated both with the exposure and the disease under study, biased risk estimates can result. However, the impact of this potential bias is rarely quantified because differences between participants and non-participants are usually unknown due to the very nature of the non-response problem. Frequently the available information from non-participants is limited to a narrow range of social and demographic variables or data on vital statistics obtainable regardless of individual agreement 3., 7., 8.. However, non-participants probably often differ in other important respects, and it is crucial to identify such differences to interpret study results correctly 3., 5., 9..

In the present study we compared demographic, behavioral, and clinical characteristics of full participants and those who only agreed to provide telephone information on a limited set of variables. Considering these specific variables we determined the expected effect of non-participation on risk estimates for acute myocardial infarction.

Section snippets

Participants and methods

This study was part of a previously described population-based case–control investigation on risk factors for myocardial infarction, the EPIcardis study 10., 11., assessing persons older than 39 years. Cases and controls were permanent residents in the catchments area of Hospital de S. João, Porto. All myocardial infarction patients living in the defined area are admitted to the cardiology wards of the hospital or referred to the cardiology department if for any reason the initial acute care is

Results

Case participation was 99.0%. Control participation was 70.0%; lower in women (66.3%) than in men (74.7%), p < 0.001. The characteristics of participants and non-participants are presented in Table 1. Regardless of gender, drinkers were significantly more common among participants. Female participants were significantly more likely to be smokers or former smokers. The mean daily number of cigarettes smoked was similar in participant and non-participant smokers (15.4 ± 11.2 vs. 16.0 ± 8.5 in

Discussion

In our sample we observed a lower proportion of participation among women, the elderly, and those who use health services more often, as indicated by the number of medical visits and health checkups in the previous year. Drinking was significantly more frequent among all participants as was smoking among female participants. However, the risk of myocardial infarction, estimated as the odds ratios for strata of the variables evaluated, were similar either including or excluding the

References (22)

  • R.N. Forthofer

    Investigation of nonresponse bias in NHANES II

    Am J Epidemiol

    (1983)
  • Cited by (0)

    This study was supported by a grant from FCT (PRAXIS/2/2.1/SAU/1332/1995).

    View full text