Background: Somatizing patients have maladaptive and increased rates of medical care utilization. If there were a way of routinely identifying such patients, one that did not require intensive, case-by-case review, they could be targeted for specific interventions to improve their use of medical care.
Objective: We sought to identify patterns of medical care utilization that would distinguish somatizing and nonsomatizing medical outpatients with acceptable sensitivity and specificity.
Design: Subjects completed questionnaires assessing somatization and sociodemographic characteristics. Their medical care utilization was obtained for the 12 months preceding the index visit. We then used multivariable logistic regression and recursive partitioning to identify patients with a provisional diagnosis of somatoform disorder. These exploratory models used various patterns of medical care utilization and sociodemographic characteristics as the independent variables.
Subjects: We studied consecutive adults attending 2 primary care practices on randomly chosen days.
Measures: The provisional diagnosis of a somatoform disorder was assessed with a 15-item self-report questionnaire. The number of primary care visits, specialty visits, mental health visits, emergency visits, and inpatient and outpatient costs were obtained for the 12 months preceding the index visit from our hospital's automated medical records, which also provided a rating of aggregate medical morbidity. Self-reported utilization outside our hospital system was obtained from a subsample of patients.
Results: Complete data were obtained on 1440 patients. Somatizing patients had more specialty care than primary care visits, higher outpatient than inpatient costs, and more emergency visits than nonsomatizing patients. A regression model containing 7 measures of utilization and 4 sociodemographic characteristics distinguished somatizing and nonsomatizing patients with a c-statistic = 0.73. Recursive partitioning identified 10 terminal nodes with a very high specificity (99%) but a very low sensitivity (15%).
Conclusions: We identified 7 discrete patterns of medical care utilization that distinguished somatizing and nonsomatizing patients. However, they did so with only modest specificity and sensitivity. This algorithm might be used effectively as the first step in a 2-step screening procedure whose second step would entail more intensive screening or individual, case-by-case review to identify somatizing patients in primary care practice.