Diagnostic judgments of nurse practitioners providing primary gynecologic care

Abstract

To determine the accuracy of experienced nurse practitioners' judgments of the probability of chlamydial infection of the cervix, to identify the clinical factors ("cues") related to the judgments, and to discern likely sources of judgment error. Cross-sectional study with prospective data collection. Urban hospital-based clinic. 492 nonpregnant women receiving primary gynecologic care. Four nurse practitioners recorded clinical data, tested women for chlamydial infection, and judged the probability of chlamydial infection using six categories: less than 1%, 1-4%, 5-9%, 10-24%, 25-50%, and greater than 50%. Chlamydial infection was detected by immunofluorescent assay in 31 (6%) of the 492 women. Although the median probability judgment was 5-9%, judgments were only weakly related (p = 0.08) to actual rates of infection. In a multivariate analysis, eight clinical cues were independently (p less than 0.05) related to nurse practitioners' probability judgments: age less than 20 years; past chlamydial or gonococcal infection; new sex partner; partner with suspected genital infection; genito-urinary symptoms; cervicitis, purulent vaginal discharge; and malodorous vaginal discharge. A linear model based on the eight cues, weighted according to their regression coefficients, predicted chlamydial infection more accurately than did the nurse practitioners' actual judgments (ROC curve areas 0.69 vs. 0.58, respectively; p less than 0.05). However, only two of the eight cues (age less than 20 years and purulent vaginal discharge) were actually related to chlamydial infection in a second multivariate model; this model bad accuracy similar to that of an empirically derived prediction rule (ROC curve areas 0.77 and 0.80, p = 0.27). Nurse practitioners were often inaccurate in their diagnostic judgments. Our analyses suggest that this inaccuracy stemmed from both the inconsistent use of clinical cues and the use of cues that were not related to chlamydial infection. Therefore, interventions such as algorithms that promote consistency and accuracy in diagnostic use of relevant cues would be likely to improve their diagnostic judgments.