Screening for high utilizing somatizing patients using a prediction rule derived from the management information system of an HMO - A preliminary study

BACKGROUND. Somatization is a common, costly problem with great morbidity, but there has been no effective screening method to identify these patients and target them for treatment. OBJECTIVES. We tested a hypothesis that we could identify high utilizing so matizing patients from a management information system (MIS) by total numbe r of visits and what we termed "somatization potential," the percentage of visits for which ICD-9 primary diagnosis codes represented disorders in the musculoskeletal, nervous, or gastrointestinal systems or ill-defined compl aints. METHODS. We identified 883 high users from the MIS of a large staff model H MO as those having six or more visits during the year studied (65th percent ile). A physician rater, without knowledge of hypotheses and predictors, th en reviewed the medical records of these patients and identified somatizing patients (n = 122) and nonsomatizing patients (n = 761). In two-thirds of the population (the derivation set), we used logistic regression to refine our hypothesis and identify predictors of somatization available from the M IS: demographic data, all medical encounters, and primary diagnoses made by usual care physicians (ICD-9 codes). We then tested our prediction model i n the remaining one-third of the population (the validation set) to validat e its usefulness. RESULTS. The derivation set contained the following significant correlates of somatization: gender, total number of visits, and percent of visits with somatization potential. The c-statistic, equivalent to the area under the ROC curve, was 0.90. In the validation set, the explanatory power was less with a still impressive c-statistic of 0.78. A predicted probability of 0.0 4 identified almost all somatizers, whereas a predicted probability of 0.40 identified about half of all somatizers but produced few false positives. CONCLUSIONS. We have developed and validated a prediction model from the MI S that helps to distinguish chronic somatizing patients from other high uti lizing patients. Our method requires corroboration but carries the promise of providing clinicians and health plan directors with an inexpensive, simp le approach for identifying the common somatizing patient and, in turn, tar geting them for treatment. The screener does not require clinicians' time.