Medical Decision Making

Purpose. To investigate the expected value of partial perfect information (EVPPI) and the research decisions it can address. Methods. Expected value of information (EVI) analysis assesses the expected gain in net benefit from further research. Where the expected value of perfect information (EVPI) exceeds the costs of additional research, EVPPI can be used to identify parameters that contribute most to the EVPI and parameters with no EVPPI that may be disregarded as targets for further research. Recently, it was noted that parameters with low EVPPI for a one-off research design may be associated with high EVPPI when considered as part of a sequential design. This article examines the characteristics and role of conditional and sequential EVPPI in EVI analysis. Results. The calculation of EVPPI is demonstrated for single parameters, groups of parameters, and conditional and sequential EVPPI. Conditional EVPPI is the value of perfect information about one parameter, conditional on having obtained perfect information about another. Sequential EVPPI is the value of perfect information for a sequential research design to investigate first one parameter, then another. Conditional EVPPI differs from the individual EVPPI for a single parameter. Sequential EVPPI includes elements from the joint EVPPI for the parameters and the EVPPI for the first parameter in sequence. Sequential designs allow abandonment of research on the second parameter on the basis of additional information obtained on the first. Conclusions. The research decision space addressed by EVI analyses can be widened by incorporating sequential EVPPI to assess sequential research designs.

Cost-effectiveness analyses (CEA) may be undertaken alongside cluster randomized trials (CRTs) where randomization is at the level of the cluster (for example, the hospital or primary care provider) rather than the individual. Costs (and outcomes) within clusters may be correlated so that the assumption made by standard bivariate regression models, that observations are independent, is incorrect. This study develops a flexible modeling framework to acknowledge the clustering in CEA that use CRTs. The authors extend previous Bayesian bivariate models for CEA of multicenter trials to recognize the specific form of clustering in CRTs. They develop new Bayesian hierarchical models (BHMs) that allow mean costs and outcomes, and also variances, to differ across clusters. They illustrate how each model can be applied using data from a large (1732 cases, 70 primary care providers) CRT evaluating alternative interventions for reducing postnatal depression. The analyses compare cost-effectiveness estimates from BHMs with standard bivariate regression models that ignore the data hierarchy. The BHMs show high levels of cost heterogeneity across clusters (intracluster correlation coefficient, 0.17). Compared with standard regression models, the BHMs yield substantially increased uncertainty surrounding the cost-effectiveness estimates, and altered point estimates. The authors conclude that ignoring clustering can lead to incorrect inferences. The BHMs that they present offer a flexible modeling framework that can be applied more generally to CEA that use CRTs.

Background. Adherence with antihypertensive and lipid-lowering therapy is poor, resulting in an almost 2-fold increase in hospitalization. Treatment side effects, cost, and complexity are common reasons for nonadherence, and physicians are often unaware of these potentially modifiable problems. Objective. To determine if a cardiovascular medication tracking and nonadherence alert system, incorporated into a computerized health record system, would increase drug profile review by primary care physicians, increase the likelihood of therapy change, and improve adherence with antihypertensive and lipid-lowering drugs. Methods. There were 2293 primary care patients prescribed lipid-lowering or antihypertensive drugs who were randomized to the adherence tracking and alert system or active medication list alone to determine if the intervention increased drug profile review, changes in cardiovascular drug treatment, and refill adherence in the first 6 months. An intention to treat analysis was conducted using generalized estimating equations to account for clustering within physician. Results. Overall, medication adherence was below 80% for 36.3% of patients using lipid-lowering drugs and 40.8% of patients using antihypertensives at the start of the trial. There was a significant increase in drug profile review in the intervention compared to the control group (44.5% v. 35.5%; P < 0.001), a nonsignificant increase in drug discontinuations due to side effects (2.3% v. 2.0%; P = 0.61), and a reduction in therapy increases (28.5% v. 29.1%; P = 0.86). There was no significant change in refill adherence after 6 months of follow-up. Conclusion. An adherence tracking and alert system increases drug review but not therapy changes or adherence in prevalent users of cardiovascular drug treatment. Targeting incident users where adverse treatment effects are more common and combining adherence tracking and alert tools with motivational interventions provided by multidisciplinary primary care teams may improve the effectiveness of the intervention.

Objectives. Having better predictors of chlamydia infection may improve health care providers’ decisions about when to provide testing for Chlamydia trachomatis (Ct). Adolescents’ probability judgments of significant life events in the next year and by age 20 y have shown promising validity, being significantly correlated with subsequent self-reports of having experienced these events. Here, the authors examine whether female adolescents’ probability judgments of having chlamydia were correlated with the objective outcome of a Ct polymerase chain reaction assay. Methods. Three hundred sexually active female adolescents were recruited from urban health care clinics in Pittsburgh. They assessed ‘‘the percent chance that you have chlamydia right now,’’ then answered questions about their demographic background and sexual history. Subsequently, the authors tested for Ct infection using a self-administered introital swab. Results. Adolescents’ probability judgments of having chlamydia ‘‘right now’’ were correlated with whether they tested positive for Ct infection, even after controlling for demographic variables and sexual history. This result held when probability judgments were dichotomized in terms of whether adolescents had assigned a zero or nonzero probability. Adolescents’ mean probability judgment was less than their infection rate, indicating that, on average, they underestimated their actual risk. Conclusions. Adolescents can tell whether they are at increased risk for chlamydia but may need better information about its absolute magnitude. Eliciting adolescents’ probability judgments of having chlamydia can add value to clinical decision making.

Parameter uncertainty, patient heterogeneity, and stochastic uncertainty of outcomes are increasingly important concepts in medical decision models. The purpose of this study is to demonstrate the various methods to analyze uncertainty and patient heterogeneity in a decision model. The authors distinguish various purposes of medical decision modeling, serving various stakeholders. Differences and analogies between the analyses are pointed out, as well as practical issues. The analyses are demonstrated with an example comparing imaging tests for patients with chest pain. For complicated analyses step-by-step algorithms are provided. The focus is on Monte Carlo simulation and value of information analysis. Increasing model complexity is a major challenge for probabilistic sensitivity analysis and value of information analysis. The authors discuss nested analyses that are required in patient-level models, and in nonlinear models for analyses of partial value of information analysis.

Background. Most patients receiving warfarin are managed in outpatient office settings or anticoagulation clinics that require frequent visits for monitoring. Objective. To measure the amount and value of time required of patients for chronic anticoagulation therapy with warfarin. Design/Participants. Prospective observation of a cohort of adult patients treated at a university-based anticoagulation program. Measurements. Participants completed a questionnaire and a prospective diary of the time required for 1 visit to the anticoagulation clinic, including travel, waiting, and the clinic visit. The authors reviewed subjects’ medical records to obtain additional information, including the frequency of visits to the anticoagulation clinic. They used the human capital method to estimate the value of time. Results. Eighty-five subjects completed the study. The mean (median) total time per visit was 147 minutes (123). Subjects averaged 15 visits per year (14) and spent 39.0 hours (29.3) per year on their visits. Other anticoagulation-related activities, such as communication with providers, pharmacy trips, and extra time preparing food, added an average of 52.7 hours (19.0) per year. The mean annual value of patient time spent traveling, waiting, and attending anticoagulation visits was $707 (median $591). The mean annual value when also including other anticoagulation-related activities was $1799 (median $1132). Conclusions. The time required of patients for anticoagulation visits was considerable, averaging approximately 2.5 hours per visit and almost 40 hours per year. Methods for reducing patient time requirements, such as home-based testing, could reduce costs for patients, employers, and companions.

The Food and Drug Administration (FDA) is considering approval of an over-the-counter, rapid HIV test for home use. To support its decision, the FDA seeks evidence of the test’s performance. It has asked the manufacturer to conduct field studies of the test’s sensitivity and specificity when employed by untrained users. In this article, the authors argue that additional information should be sought to evaluate the prevalence of undetected HIV in the end-user population. The analytic framework produces the elementary but counterintuitive finding that the performance of the home HIV test— measured in terms of its ability to correctly detect the presence and absence of HIV infection among the people who purchase it—depends critically on the manufacturer’s retail price. This finding has profound implications for the FDA’s approval process.

Introduction. The implementation of mass colorectal cancer (CRC) screening is a public health priority. Population participation is fundamental for the success of CRC screening as for any cancer screening program. The preferences of the population may influence their likelihood of participation. Objectives. The authors sought to elicit population preferences for CRC screening test characteristics to improve the design of CRC screening campaigns. Methods. A discrete choice experiment was used. Questionnaires were compiled with a set of pairs of hypothetical CRC screening scenarios. The survey was conducted by mail from June 2006 to October 2006 on a representative sample of 2000 inhabitants, aged 50 to 74 years from the northwest of France, who were randomly selected from electoral lists. Questionnaires were sent to 2000 individuals, each of whom made 3 or 4 discrete choices between hypothetical tests that differed in 7 attributes: how screening is offered, process, sensitivity, rate of unnecessary colonoscopy, expected mortality reduction, method of screening test result transmission, and cost. Results. Complete responses were received from 656 individuals (32.8%). The attributes that influenced population preferences included expected mortality reduction, sensitivity, cost, and process. Participants from high social classes were particularly influenced by sensitivity. Conclusions. The results demonstrate that the discrete choice experiment provides information on patient preferences for CRC screening: improving screening program effectiveness, for instance, by improving test sensitivity (the most valued attribute) would increase satisfaction among the general population with regard to CRC screening programs. Additional studies are required to study how patient preferences actually affect adherence to regular screening programs.

Background. Patients’ assessment of satisfaction with care, quality of care, and outcomes has become a central issue in patient-centered prostate cancer (PCa) care. We sought to analyze the association between patient-reported satisfaction with care and health-related quality of life (HRQoL) in newly diagnosed PCa patients. Methods. Prospective cohort design was used to recruit 590 newly diagnosed PCa patients from an urban academic hospital and a VA hospital. Participants completed satisfaction with care (CSQ-8) and HRQoL (SF-36 and UCLA-PCI) surveys prior to treatment and at 3, 6, 12, and 24 months’ follow-up. Cross-lagged analysis was used to ascertain the causal direction between satisfaction with care and HRQoL. Propensity scores were used to adjust for potential selection bias between treatment groups. Linear mixed models were used to analyze the association between satisfaction with care, process of care (treatment), and outcomes (generic and prostate-specific HRQoL) after adjusting for covariates. Results. Cross-lagged correlation results are consistent with a cause-effect association between HRQoL and satisfaction with care. After controlling for clinical and demographic covariates, radical prostatectomy (RP) treatment was associated with higher satisfaction with care (odds ratio [OR], 7.9; P = 0.043). Improved generic and prostate-specific HRQoL were associated with higher satisfaction with care, after adjusting for demographic and clinical covariates. Conclusion. Satisfaction with care appears to be associated with process of care and outcomes of care. Assessment of satisfaction with care is useful for evaluating the quality of PCa care. Satisfaction with care is an important arena in cancer outcomes research, whose full potential remains unexploited.

Systematic reviews of healthcare disparities suggest that clinicians’ diagnostic and therapeutic decision making varies by clinically irrelevant characteristics, such as patient race, and that this variation may contribute to healthcare disparities. However, there is little understanding of the particular features of the healthcare setting under which clinicians are most likely to be inappropriately influenced by these characteristics. This study delineates several hypotheses to stimulate future research in this area. It is posited that healthcare settings in which providers experience high levels of cognitive load will increase the likelihood of racial disparities via 2 pathways. First, providers who experience higher levels of cognitive load are hypothesized to make poorer medical decisions and provide poorer care for all patients, due to lower levels of controlled processing (H1). Second, under greater levels of cognitive load, it is hypothesized that healthcare providers’ medical decisions and interpersonal behaviors will be more likely to be influenced by racial stereotypes, leading to poorer processes and outcomes of care for racial minority patients (H2). It is further hypothesized that certain characteristics of healthcare settings will result in higher levels of cognitive load experienced by providers (H3). Finally, it is hypothesized that minority patients will be disproportionately likely to be treated in healthcare settings in which providers experience greater levels of cognitive load (H4a), which will result in racial disparities due to lower levels of controlled processing by providers (H4b) and the influence of racial stereotypes (H4c).The study concludes with implications for research and practice that flow from this framework.

Objectives. To test the hypothesis that nearest-neighbor analysis adds to logistic regression in the early diagnosis of late-onset neonatal sepsis. Design. The authors tested methods to make the early diagnosis of neonatal sepsis using continuous physiological monitoring of heart rate characteristics and intermittent measurements of laboratory values. First, the hypothesis that nearest-neighbor analysis makes reasonable predictions about neonatal sepsis with performance comparable to an existing logistic regression model was tested. The most parsimonious model was systematically developed by excluding the least efficacious clinical data. Second, the authors tested the hypothesis that a combined nearest-neighbor and logistic regression model gives an outcome prediction that is more plausible than either model alone. Training and test data sets of heart rate characteristics and laboratory test results over a 4-y period were used to create and test predictive models. Measurements. Nearest-neighbor, regression, and combination models were evaluated for discrimination using receiver-operating characteristic areas and for fit using the Wald statistic. Results. Both nearest-neighbor and regression models using heart rate characteristics and available laboratory test results were significantly associated with imminent sepsis, and each kind of model added independent information to the other. The best predictive strategy employed both kinds of models. Conclusion. The authors propose nearest-neighbor analysis in addition to regression in the early diagnosis of subacute, potentially catastrophic illnesses such as neonatal sepsis, and they recommend it as an approach to the general problem of predicting a clinical event from a multivariable data set.

Objective. A difficulty with applying decision analysis at the bedside is that it generally requires computer software for the calculations, which may render the method impractical. The purpose of this study was to illustrate the feasibility of developing a regression model that approximates the results from a published decision-analytic model for prostate cancer and permits bedside generation of personalized decision-analytic recommendations with a paper nomogram. Methods. The authors used the example of radical prostatectomy v. watchful waiting for patients with early-stage prostate cancer. First, they took a published decision analysis and generated recommendations using simulated data where patient baseline factors and preference scores for health states were systematically varied. Multivariable logistic regression was used to identify the parameters with strong associations with the recommendation. A reduced model was fit that excluded other preference scores except for watchful waiting. They compared the recommended management predictive accuracies from the full v. reduced model at the individual patient level for 63 men from another published study. Discrimination was assessed using receiver operating characteristic (ROC) curve analysis. A nomogram was constructed from the covariates in the reduced model. Results. The reduced logistic regression model predicted the recommendations accurately for the 63 patients, with an area under the ROC curve of 0.92. Discrimination was excellent as demonstrated by histograms. Conclusions. The authors demonstrated that logistic regression modeling allows accurate reproduction of decision-analytic recommendations with simplified calculations, which can be accomplished using a graphic nomogram. This approach should facilitate clinical decision analysis at the bedside.

Background. Aircraft noise disturbs sleep and impairs recuperation. Authorities plan to expand Frankfurt airport. Objective. To quantitatively assess the effects of a traffic curfew (11 PM to 5 AM) at Frankfurt Airport on sleep structure. Design. Experimental sleep study; polysomnography for 13 consecutive nights. Setting. Sleep laboratory. Subjects. 128 healthy subjects, mean age (SD) 38 (13) years, range 19 to 65, 59% female. Intervention. Exposure to aircraft noise via loudspeakers. Measurements. A 6-state Markov state transition sleep model was used to simulate 3 noise scenarios with first-order Monte Carlo simulations: 1) 2005 traffic at Frankfurt Airport, 2) as simulation 1 but flights between 11 PM and 5 AM cancelled, and 3) as simulation 2, with flights between 11 PM and 5 AM from simulation 1 rescheduled to periods before 11 PM and after 5 AM. Probabilities for transitions between sleep stages were estimated with autoregressive multinomial logistic regression. Results. Compared to a night without curfew, models indicate small improvements in sleep structure in nights with curfew, even if all traffic is rescheduled to periods before and after the curfew period. For those who go to bed before 10:30 PM or after 1 AM, this benefit is likely to be offset by the expected increase of air traffic during late evening and early morning hours. Limitations. Limited ecologic validity due to laboratory setting and subject sample. Conclusions. According to the decision analysis, it is unlikely that the proposed curfew at Frankfurt Airport substantially benefits sleep structure. Extensions of the model could be used to evaluate or propose alternative air traffic regulation strategies for Frankfurt Airport.