Combining unequal variance signal detection theory with the health belief model to optimize shared decision making in tinnitus patients: part 1-model development

Abstract

INTRODUCTION: The results from different Cochrane studies justify considerable professional equipoise concerning different treatment options for tinnitus. In case of professional equipoise, Shared Decision Making (SDM) is an indispensable tool for guiding patients to the intervention that best fits their needs. To improve SDM we developed a method to assess the accuracy and utility of decisions made by tinnitus patients when freely choosing between different treatment options during their patient journey. The different treatment options were audiological care and psychosocial counseling. METHODS: We developed a statistical model by combining Signal Detection Theory (SDT) with the Health Belief Model (HBM). HBM states that perceived severity of an illness is strongly related to sick-role behavior. As proxies for perceived severity, we selected hearing loss and Tinnitus Handicap Inventory (THI) score at baseline. Next, we used these proxies as predictors in linear regression models based on SDT to determine the likelihood ratio of true positive decisions (choosing a treatment option and experiencing an improvement of more than 7 points in THI-score) and false positive decisions (choosing a treatment option and experiencing an improvement of less than 7 points in THI-score) for audiological care and psychosocial counseling, respectively. Data was gathered in a prospective cohort of 145 adults referred for tinnitus care to an outpatient audiology clinic in the Netherlands. The participants were asked to decide freely on uptake of audiological care (provision of hearing aids with or without a sound generator) and uptake of psychosocial counseling. Logistic regression with Bayesian inference was used to determine the cumulative distribution functions and the probability density functions of true positive decisions and false positive decisions as function of hearing loss and baseline THI-score for both treatment options, respectively. With the cumulative distribution functions, we determined the accuracy of the decisions. With the probability density functions we calculated the likelihood ratios of true positive decisions versus false positive decisions as function of hearing loss and baseline THI-score. These likelihood ratio functions allow assessment of the utility of the decisions by relating a decision criterion to perceived benefits and costs. RESULTS: Baseline THI-score drives decisions about psychosocial counseling and hearing loss drives decisions about audiological care. Decisions about psychosocial counseling are more accurate than decisions about audiological care. Both decisions have a low accuracy (0.255 for audiological care and - 0.429 for psychosocial counseling), however. For decisions about audiological care the unbiased decision criterion is 37 dB(HL), meaning that a lenient decision criterion (likelihood ratio < 1) is adopted by patients with a hearing loss below 37 dB and a strict criterion (likelihood ratio > 1) by patients with a hearing loss exceeding 37 dB. For psychosocial counseling uptake the decision criterion is always strict, regardless of baseline THI-score. The distributions of the populations, that do and do not experience a clinically important change in THI-score, have unequal variances for psychosocial counseling, while they have almost equal variances for audiological care. DISCUSSION: Combining SDT and HBM can help assess accuracy and utility of patient decisions and thus may provide valuable information that can help to improve SDM by combining patient related outcome measures, decision drivers, and perceived benefits and costs of a treatment.