Estimating the prevalence of medically unexplained symptoms from primary care records.

OBJECTIVES

To develop models to estimate the likely prevalence of medically unexplained symptoms (MUS) and severe MUS in a primary care practice from existing patient electronic records collected in the previous 2 years for secondary prevention and commissioning of psychological treatment.

STUDY DESIGN

Cross-sectional survey comparing general practitioners' (GPs) assessment of the presence or absence of MUS and severe MUS with clinical, demographic and service use variables associated with MUS or functional somatic syndromes from previous research in the patient's routine electronic record over the previous 2 years.

METHODS

Seventeen GPs from eight practices identified cases of MUS and severe MUS in 828 consecutive consulters in primary care. Models of variables associated with MUS and severe MUS were constructed using multivariate multilevel logistic regression. The predictive validity of the final models was tested, comparing predicted with observed data and expected prevalence rates from the literature.

RESULTS

Models to predict MUS and severe MUS had areas under the receiver operating characteristic curve of 0.70 [95% confidence interval (CI) 0.65-0.74] and 0.76 (95% CI 0.70-0.82), respectively. Both models showed adequate goodness of fit with observed data, and had good predictive validity compared with the expected prevalence of MUS, severe MUS, and anxiety or depression.

CONCLUSION

Models to predict the prevalence of MUS and severe MUS from routine practice records for commissioning purposes were successfully developed, but they require independent validation before general use. The sensitivity of these models was too low for use in clinical screening.