The influence of first impressions on the diagnosis of early cancers by GPs

Talk Code: 

The problem

Late diagnosis is a major contributor to poor cancer survival rates. One suggested reason is GP delay. The study aimed to investigate whether initial diagnostic impressions determine GPs’ final diagnoses of early cancers.

The approach

We constructed 3 detailed patient scenarios, where cancer was a possible diagnosis (colorectal, lung, and myeloma), and which did not contain ‘red flags’. Each patient presented twice, the second time with non-improving or additional symptoms. In each scenario, a more common diagnosis partially explained the symptoms.GPs participated remotely over the Internet, while on the phone with a researcher. After reading a patient description and the reason for encounter (RfE), they could elicit further information as they saw fit. The researcher answered the GPs’ questions from a pre-determined information pool. At each patient visit, participants recorded their diagnoses (working and differential). In 2 of the 3 scenarios, we asked participants to ‘think aloud’.Using the think-aloud protocols, we identified the participants’ first impressions, based on their very first utterances, after they had read the patient description and RfE, and before asking any further questions. Two independent raters coded these utterances as either acknowledging the possibility of cancer or not. We used multilevel logistic regression to measure the influence of first impressions on final diagnosis (i.e., whether cancer was given as the working diagnosis or in the differential). We performed mediation analysis to measure whether this influence was mediated by information search.


Ninety GPs participated. Across scenarios and visits, cancer was diagnosed 51% of the time. Mean inter-rater agreement on first impressions was substantial (Kappa=0.70). First impressions that did not acknowledge the possibility of cancer were associated with fewer questions (? -2.34, [-0.75 to -3.94], P =0.004), and with reduced odds of a final cancer diagnosis (OR 0.26 [0.15 to 0.48], P <0.001) and urgent referral (OR 0.43, [0.22 to 0.84], P =0.013). The effect of first impressions on final diagnosis was only partially mediated by information search (number of questions asked). More experienced GPs were less likely initially to acknowledge the possibility of cancer (OR 0.93 [0.91 to 0.97], P <0.001), and somewhat less likely to refer urgently (0.98 [0.96 to 1.00], P=0.056).Consequence: This is the first study to provide empirical evidence that first impressions, before any information gathering takes place, can determine final diagnosis, irrespective of the intervening information search. This can be problematic in cases without strong diagnostic features, such as early cancers, and suggests that attempts to improve diagnostic judgements should target the early stages of the diagnostic process. GPs with more years in general practice are less likely to consider cancer in the absence of ‘red flags’, possibly by relying more on base rates.


  • Olga Kostopoulou
  • Miro Sirota
  • Tom Round