Background:
Misdiagnosis occurs in about 5% of outpatients, and in 10% to 35% of emergency room (ER) patients, sometimes with devastating medical and economic consequences. Nowadays, computerized diagnostic decision support programs (CDDS) exist, which suggest differential diagnoses (DDx) to physicians and thus have potential to improve diagnoses and hence, outcomes of patient care. The effects of such CDDS in 'real-world' ER settings are unknown. Controlled clinical trials investigating their effectiveness and safety are absent. In addition, most available CDDS are overcautious and suggest a wide variety of diagnostic options, likely increasing diagnostic resource consumption.
Objectives:
With this project, the investigators aim to understand the intended and unintended consequences of CDDS use by physicians on diagnostic quality and workflow in emergency medicine
- on the micro-level, how CDDS affect diagnostic quality by physicians in individual emergency patients.
- on the meso-level, how CDDS affect the diagnostic workflow in emergency departments.
- on the macro-level, the economic and educational impact of CDDS utilization in ERs
Outcomes: Details given below
Design:
Cross sectional, multi-center, four-period cross-over controlled cluster-randomized trial. Four ER sites will randomly be allocated to one of two sequences with alternating intervention and control periods (ABAB vs. BABA) with each period lasting for two months. Recruitment will target 74 patients per period and cluster and 1'184 patients total.
Inclusion / Exclusion Criteria: Details given below
Intervention period: Details given below
Control period: Details given below.
Measurements and procedures:
For the primary outcome, data will be extracted from the electronic health records (i.e. ER diagnosis, intensive care unit admission or diagnosis after 14d if patients are still hospitalized). Additionally, patients and their general practitioner will be contacted via telephone by study nurses after 14d of study inclusion in order to collect information about patients' current diagnoses, and re-visits or hospitalization related to the initial ER visit. Data for secondary endpoints will be retrieved from the routinely collected data in the electronic health record system (e.g mortality, time to ER diagnosis, resource consumption). Additionally, interviews and focus groups with physicians will be performed to investigate diagnostic workflow changes, physician confidence and other process outcomes.
Statistical Analysis:
Statistical analysis will be based on multi-level general linear mixed modelling (GLMM) methods using appropriate post hoc techniques (e.g for subgroup analyses).
For the primary outcome (presence or no presence of a positive diagnostic quality risk score), a generalized linear mixed model (GLMM) with a binomial distribution family and exchangeable correlation structure will be performed. The GLMM takes into account a random effect for each site, resident and attending physician. Diagnosing resident and attending physicians are nested within sites. The condition (intervention and control) and the period (period 1 to 4) will be included as fixed factors under the assumption of equality of carry-over effects. Additionally, presenting chief complaint, patient's age, sex and comorbidity index will be added as covariates.
For all secondary endpoints, summary statistics appropriate to the distribution will be tabulated by treatment group. Analysis of secondary endpoints will parallel the primary analysis.
