[Reliability of morbidity data reported by GPs. Results of a longitudinal study in primary care].

BACKGROUND

Data on prevalence of chronic diseases are important for planning health care services. Such prevalence data are mostly based on patient self-reports, claims data, or other research data-with limited validity and reliability partially due to their cross-sectional character. Currently, only claims data of statutory health insurance offer longitudinal information. In Germany, these data show a loss of diagnoses of chronic health conditions over time. This study investigated whether there is a similar tendency of loss in the documentation of chronic diseases in data specifically collected for a longitudinal cohort study by general practitioners. In addition, the explanatory power of patient or GP characteristics regarding these losses is investigated.

PATIENTS AND METHODS

A total of 3,327 patients aged 75 years and older were recruited for the German Study on Ageing, Cognition and Dementia in Primary Care Patients (AgeCoDe). For 1,765 patients, GP diagnoses of four chronic conditions at three time points were available for a total period of 4.5 years. In order to explain the loss of chronic diagnoses, a multilevel mixed-effects logistic regression was performed.

RESULTS

Over the course of 4.5 years, 18.6% of the diagnoses of diabetes mellitus, 34.5% of the diagnoses of coronary heart disease, and 44.9% of the diagnoses of stroke disappeared in the GP documentation for the longitudinal study. The diagnosis of coronary heart disease was less often lost in men than in women. The risk of losing the diagnosis of diabetes was higher in patients who were well known by the GP for a long time. An essential part of the variance of the losses can be explained by practice (owner) effects.

CONCLUSION

Data on morbidity collected in epidemiological studies and reported by physicians should always be checked for validity and reliability. Appropriate options (e.g., an investigator collecting the data directly in the field or the comparison of the data with health insurance companies' claims data) are presented and discussed.