Using general practitioners to measure community morbidity

Randomly-selected patients drawn from randomly-selected General Practitioners (GPs) (two-stage cluster sample) were compared with a sample of the general population, who had visited a GP, selected using close approximations to standard household survey methods (area probability) of the Australian Bureau of Statistics. If GP patients drawn in this way resemble a random sample of the Australian community who have recently used GP services, then confidence should increase in this much cheaper method as a source of morbidity statistics. Interviews focused upon each person's last visit to the GP, with questions about reasons for attending, diagnoses and treatments, and various demographic items. In univariate analyses of 22 demographic items, 17 consultation items and 27 diagnoses and treatments, only five items were differently distributed between the GP patients and the area sample. Pairs of data items were also similar in the two groups. Items were examined using multidiscriminant analysis, to determine those that discriminated between the two groups and to calculate predicted group membership on the basis of these items. This analysis correctly classified only 56.7% of study subjects into their true group (GP patient or area sample) when based on items that were differently distributed between the groups, and 53.3% when all items were used, indicating that discrimination was only slightly better than chance. This result increases the confidence with which GP patients can be used to estimate levels of morbidity in the community if random selection is used to select GPs and if their patients are also randomly selected.