Cancer causes death to millions of people worldwide. Early detection of cancer in primary care can enhance patients’ chances of survival. However, detecting cancer early is no simple task. The symptoms can be vague and non-specific, and can be easily attributed to pre-existing or other, more common conditions. Attributing symptoms of bowel or ovarian cancer, such as abdominal distention and pain, to Irritable Bowel Syndrome is a frequent example. With funding from Cancer Research UK, we have been researching early cancer diagnosis for the last 5 years. When I started work in this area, I had assumed that cancer is always at the back of a GP’s mind. I found out that this was not necessarily the case.
My team and I built detailed patient cases where cancer was a possibility, and presented them to GPs on computer. We asked them to ‘think aloud’, which means, to say whatever thoughts crossed their mind, without explaining anything. We then coded the first thing that each GP said as either ‘cancer mentioned’ or ‘cancer not mentioned’, and measured the statistical association between these ‘first impressions’ and the final diagnoses. We found that if GPs mentioned cancer explicitly at the start of a case, they were much more likely to diagnose it and refer the patient at the end (after they had gathered any other information that they needed as part of their patient assessment). This study demonstrated the strong influence of ‘first impressions’ on subsequent diagnosis and management, and suggested that if a hypothesis is not considered at the start, it is unlikely to be pursued later on. You can read the full study here.
In our attempt to improve diagnosis, we obtained funding from the EU, and led the TRANSFoRm project, which included the development of a prototype Decision Support System (DSS) for diagnosis in primary care. The DSS aims to support diagnostic ‘first impressions’. It integrates with the electronic health record and works by using patient information from the record (age, sex, risk factors) and the patient’s ‘Reason for Encounter’ (RfE), i.e. why the patient has come to see the GP. Once the GP enters the RfE, the DSS presents a list of possible diagnoses, ordered by incidence, which the GP is asked to simply read. Subsequently, as the GP codes symptoms and signs into the DSS (the DSS interface facilitates coding), the ordering of the diagnoses on the list changes. Of course, the DSS never presents the GP with a single final diagnosis, as most diagnoses cannot be excluded in primary care, and the evidence available is not sufficient for this. The DSS merely aims to suggest to the GP diagnostic alternatives that he/she might not have considered, and to do this as early as possible in the consultation.
Our evaluation of the DSS prototype in a high-fidelity simulation, where GPs consulted with a range of ‘standardised patients’ (actors), found it to be effective: average diagnostic accuracy across cases (including cancers) rose from 50% to 58%. This is a statistically significant improvement, which could translate into clinically significant gains given the number of primary care consultations. The GPs found the prototype to be useful and usable, while ‘patient’ satisfaction, measured using a standardised questionnaire that the actors filled in after each consultation, was not affected by the DSS. In a piece titled “Computer helps doctors spot serious illness”, The Times newspaper reported on the study (31 January, 2017). You can read the full study here.
In summary, there is strong evidence that initial hypotheses influence final diagnosis and that supporting the initial stage of hypotheses generation by external decision support can result in more accurate and earlier diagnosis and referral of serious conditions, including cancers.