D2.1 – Most Promising Type of Support for Diagnosis

A computerised interface was designed by King’s College London (KCL) to present the scenarios to GPs and collect the data remotely over the Internet using a web tool. Participating GPs could gather information about the simulated patients by asking questions that a researcher answered (i.e., the researcher selected the appropriate answer from a list of predetermined cues and sent it to the participant’s screen in real time). Participants and researcher were in concurrent phone communication. The web tool automatically recorded all questions participants asked in sequence and the time taken.

Early diagnostic support involved presenting participants with a list of diagnostic suggestions as soon as they read some background patient information (age, sex, risk factors, current medications, past medical history) and the reason for encounter. The list remained on screen for a minimum of 20 seconds and disappeared as soon as the participants started their information gathering in order to diagnose and manage the scenario patients. This early support aimed to help participants consider more diagnostic hypotheses than they otherwise would and reduce the tendency quickly to narrow down on one or a small number of hypotheses. Participants could make the list of diagnostic suggestions appear again on their screen at any point they wished during the consultation with a patient.

Another group of participants received a list of diagnostic suggestions only after they had gathered information and had given their own diagnosis and management. This list depended on the information each participant had gathered during the consultation. It aimed to alert participants to diagnoses that they had not excluded and represented the more traditional approach to diagnostic support, where physicians first gather information as they see fit and then enter all the information into the system to receive advice about differential diagnoses. After seeing the list, participants could request further information, change their diagnosis or give the same diagnosis.

The lists of diagnostic suggestions for each scenario were derived from an existing, stand-alone, diagnostic support system for internal medicine, DxPlain (http://dxplain.org/), and were adapted to primary care by the KCL team. These lists constituted the early support lists for each scenario. The late support lists originated from the early support lists, which were narrowed down depending on the information each participant had gathered. The KCL team formulated rules that determined which diagnoses from the early support lists could reasonably be discounted, had a participant requested specific cues. Thus, late support was individualised to each participants’ information search.

The study initially took place in the UK. All study materials were translated and adapted to a Greek context by the teams at the University of Crete (UoC) and KCL. The study was repeated in Greece with a smaller sample of GPs in order to ascertain the feasibility and likely effectiveness of diagnostic support in a country with a newer and substantially different primary care system than the UK. There was significant heterogeneity between the studies in the two countries, therefore, the data were not pooled but were analysed separately.

297 GPs in the UK and 150 GPs in Greece were allocated via blocked randomisation to one of three experimental conditions: control, early support or late support. The primary outcome was diagnostic accuracy. In the UK, early support significantly improved diagnostic accuracy over control (OR 1.31 [95% CI 1.04-1.66], P=0.023), while late support did not (OR 1.10 [95% CI 0.87-1.38]). In Greece, both types of support significantly improved accuracy over control: OR 1.68 [95% CI 1.23-2.30] (P=0.001) for early support and OR 1.50 [95% CI 1.10-2.04] (P=0.011) for late support.

Reminding GPs of diagnostic possibilities to consider early on in the consultation, before they start narrowing down on a hypothesis, can significantly improve diagnostic accuracy across a wide range of diagnostic scenarios, without significantly increasing the amount of information gathered. In a sample of GPs in Greece, both early and late reminders improved diagnostic accuracy over control. Further research in Work Packages 2, 4 and 5 will thus focus on developing a DSS prototype, integrated with the EHR, that has the potential to provide early reminders in response to the reason for encounter but is also flexible enough to make diagnostic suggestions at other points during the consultation and with varying amounts of information.