Use of general practitioner computerised records to create a population based twin sample: pilot study based on Parkinson's disease

Introduction

Studies of twins are useful in investigating the role of genetics in disease. We describe a population based method of twin ascertainment in Parkinson's disease using doctors' records.

Subjects, methods, and results

All doctors in four East Anglian counties were asked to search their database for patients receiving drugs for Parkinson's disease. They were asked to confirm the diagnosis of Parkinson's disease so that we could exclude patients taking these drugs for other conditions. We contacted each patient by letter through their doctor to confirm the diagnosis and to inquire whether they were one of twins—zygosity was based initially on patients' self reports.

Overall, 149 general practice surgeries took part (44% of those contacted; estimated study population 1.53 million). Seventy per cent (1527) of patients replied to the first letter, increasing to 87% (1898) after one reminder—2182 patients in total. There was no significant difference in the mean age of respondents and non-respondents (75.8 SD (19.4) years v 77.4 (19.4) years; t =1.87, P>0.1). Forty eight respondents (27 men, 21 women) reported that they were one of twins (54 would be expected given a ratio of singleton to twin births in the United Kingdom of 1:40). There were no concordant pairs and no one was ascertained twice. Eight patients were one of monozygotic twins (7 male pairs), and 20 were one of dizygotic twins (10 were like sexed; 5 male and 5 female pairs); in 20 cases zygosity was unknown because the twin had been stillborn (6) or data were lacking (14). Assuming Parkinson's disease occurs in men and women in the ratio of 1.3:1,1 this matches the overall sex distribution of 27 male and 21 female twins. Men were slightly underrepresented among the dizygotic twins (20 pairs instead of 22.5), but the ratio was still within the 95% confidence interval of 1.3:1. The apparent excess of monozygotic male pairs was also non-significant (P=0.074). The expected proportion of monozygotic twin pairs calculated by Weinberg's rule2 was 0.29 (8/28), as was the observed proportion.

Comment

Despite the low response rate of general practitioners we have shown that twins can be identified from prescribing indices. Given that the prevalence of Parkinson's disease is 164 per 100 000 population,3 we failed to detect about 12% of cases (2525). This may reflect regional differences or affected patients may have been missed because they were not receiving treatment; some receiving treatment may not have had Parkinson's disease but essential tremor and parkinsonian syndromes, for example, and our figure will be an overestimate.

If the expected twin to singleton ratio is applied only to respondents (1861) there was no shortfall (47 twins would be expected). Twins seem not to have responded preferentially, and being a twin is unlikely to be a factor in the aetiology of Parkinson's disease. Patient response was 87% and not biased with age—the mean age of 75.8 years compares favourably with that of 75.3 years in Aberdeen.3

The reply rate of doctors was modest (44%). As only 83% of surgeries were computerised at the time of study this equates to a 53% response rate. There may have been some bias from doctors if, for example, they had better records or interesting or atypical patients, but the data are closely within the expected range so any bias from this source should be small. Our study was biased in favour of group surgeries as 99% were computerised compared with 66% of singlehanded practices. In 1997, 92% of surgeries should be computerised,4 so sampling will be more efficient. The monozygotic to dizygotic ratio in our sample is approximate because it relies on self reporting, although this has a surprisingly low error rate (5%).5

The diagnosis of Parkinson's disease in this study is presumptive. In subsequent studies all twins would need to be interviewed. Where a twin registry and record linkage are not available our method might be easier than other population based approaches.