Trends in lung cancer emergency presentation in England, 2006–2013: is there a pattern by general practice?

We aimed to describe and explain the heterogeneity in proportions of lung cancer diagnoses through emergency presentation – thereby referred to as proportions of EP - between practices in 2006–2013. First, we explored the variability in the national proportions of three types of emergency presentations over time. Then, we depicted the variation in proportions of emergency presentation by practice. Finally, we explored the association between practice characteristics and proportions of emergency presentations adjusting for variations by patient characteristics.

Figure 2 shows a series of funnel plots describing the heterogeneity in practice-specific proportions of EP for the years 2006 to 2013. It highlights where practices stand with respect to their proportions of EP of lung cancer patients in the corresponding year, by their number of new cases of lung cancer as a measure of precision of the estimates. For a given year, very few practices had proportions of EP above the upper 95% confidence limit: for example, in 2010, this included only 150 practices out of 7514 (red triangles in Fig. 2). However, in the same calendar year, half of the practices (3667) presented a maximum of three lung cancer patients, which means that setting the EP proportion at the extreme level of 0% in the 150 upper outliers would decrease the national level of EP by only 2%. Furthermore, practice-level proportions changed dramatically year on year because of the high number of practices with few patients: cumulatively, as many as 1163 General Practices in England, i.e. 14.6% of the total number of General Practices, fell above the upper limit of the funnel plots at least once between 2006 and 2013.

Explanatory and confirmatory factor analyses reduced the practice-level datasets to only two or three factors for each of the different data sources. The factors could be labelled “Trust and confidence in the nurse” and “Trust and confidence in the GP” from the GPPS data, “Diagnosis of cancer”, “Two-Week Wait referrals” and “Use of screening” from the PP data, and “COPD” and “Smoking services” from the QOF data. The few factors retained were not predictive of emergency presentation (Odds Ratio, OR, close to 1, data not shown).

Additional file 3: Table S3 indicates that there was no evidence of practice-level clustering of EP. In fact, the VPC was extremely small in most cases, around 1%, and the AIC also favoured models without random intercept. Consequently, all conclusions were based on logistic regression models. All variable selection techniques led to similar conclusions that most practice-specific variables were not associated to EP. The stage-specific analysis identified a few practice characteristics which were weakly associated with lung cancer EP. These were: getting through to the practice on phone, ability to see a doctor in next two days or more than two days after last tried, how good the doctor is at asking about symptoms, how good the nurse is at giving enough time. The odds ratios (ORs) between each predictor and emergency presentation were not significant and with wide confidence intervals (see Additional file 4: Table S4). The addition of variables to the model did not improve its predictive performance. In Additional file 5: Figure S1 we provide the ROC curves and the corresponding C-index, which illustrate the prediction ability of these models. Comparing panels A and B, the practice-level variables did not improve the predictive performance of the models: C-index increased by only 1%, which was irrelevant for practical purposes. We obtained the same conclusions using the Brier score, which represents an alternative measure of predictive accuracy.

Although the proportion of lung cancer emergency presentations decreased slightly over the time period examined, the number of patients diagnosed every year increased, resulting in a stable number of emergency cases per year (Table 1). GPs’ role as gate keepers for secondary care aims to enhance the appropriateness of setting for patient care and to reduce unnecessary pressure on secondary care [28]. Nonetheless our results suggest that two-thirds of patients presenting as EP by-pass primary care. A better understanding is needed about the motivations and the previous primary care pathway that led these patients to access care via A&E. This high proportion, combined with (a) the concomitant decrease in GP-led EP proportion and increase in patient-led EP proportion, and (b) the increasing proportion of EP from A&E of another healthcare provider, is noteworthy. Our results cannot be attributed to any change in definition of these sub-types of EP, because definitions did not change between 2006 and 2013.

We also show that the least deprived patients are more likely to be referred by their GP to A&E compared to the most deprived. The extent to which this might be driven by more frequent GP attendance, higher levels of health literacy or other factors could not be explored in this analysis.

We found that the high proportion of EP is not the result of a few practices with very abnormal patterns of EP. Rather, it is the combined effect of the great majority of practices with proportions of EP around an already high national average. An illustration of the extent of this problem is that, to reduce the national average of EP from 37.6 to 30% in 2010, 662 practices with observed highest proportions of EP (9% of all practices diagnosing lung cancer patients that year) would need to have a proportion of EP of 0%. This numerical example illustrates that a targeted intervention on a few practices [29] may have little effect on the national proportions. Furthermore, the targeted practices would change every year.

Previous research identifies practice characteristics associated with increased EP. These include poorer access to general practice, measured as the proportion of patients who were able to obtain an appointment on their last attempt [30], and lower proportions of patients who had confidence and trust in their doctor [31], and discontinuity in consultation [32]. However, none of these analyses include an evaluation of the predictive performance of the selected model and variables, which reduces the utility of the findings. Our research suggests that none of the GP practice characteristics available for national-level analysis satisfactorily predict EP. Our results are in line with a review of 22 studies investigating EP in lung and colorectal cancer patients, concluding that no study found clear evidence between primary care factors and EP [33]. Similarly in 2001, it was established that emergency admission of colorectal cancer was not associated with present aspects of primary health care organization [34].

A recent review on the evidence about emergency presentations [35] highlights that, although the mechanisms leading to EPs are not fully understood, there is still a substantial proportion of avoidable emergency presentations. Avoidable EPs are hypothesised to result from several successive or independent “omissions”, on the part of the patients and GPs, with respect to their actions towards signs and symptoms of cancer [36]. Patients may lack knowledge of cancer symptoms, or delay seeking health advice or investigations [37]. Nation-wide campaigns such as “Be clear on cancer” aim to tackle these causes of delay. In addition, primary care practitioners may overlook cancer symptoms, delay tests, investigations or referrals to secondary care [38]. This may in part be associated with GPs and patients prioritising other complaints: some EPs are the result of the combination of several factors, including the presence of other diseases [39].

To our knowledge, this study is the most comprehensive analysis on the associations between the general practice characteristics and EP. We linked administrative and survey data to records of cancer patients from the well-established population-based Cancer Registry for England. We adopted different analytical approaches to investigate those associations, i.e. exploratory and confirmatory structural equation modelling approach, as well as two model selection strategies. We finally looked at the predictive performance of the selected models and factors.

However our research is limited by our lack of information on the extent of primary care involvement in ‘patient-led’ attendances. This meant that we were unable to estimate the proportion of patients coded as patient-led EP who were sent to A&E by their GPs. These patients may explain some of the rise in patient-led EP, and the sudden increase from 2011 in proportions of patients referred to A&E via “the A&E department of another healthcare provider”. Furthermore, previous research which linked CPRD (Clinical Practice Data Link) to cancer registrations for colorectal cancer [40] showed that although primary care use and access was similar between patients with and without EP of their colon cancer, EP patients were less likely to have red-flag symptoms recorded in primary care in the year prior to the diagnosis.

GP practices only see a limited number of lung cancer patients every year, leading to high variability in their proportions of EP. Nonetheless the methods used in this paper to detect associations have a good power since the sample sizes remain a lot larger than the number of parameters, and in all cases the associations exhibited very low significance level. Furthermore, even when all types of emergency presentations are studied, there is limited evidence for association [32].

Finally, anonymised GPPS information does not allow us to study the experience of patients who by-pass their GPs compared with those who do not. Moreover, we do not know the extent to which communication barriers between GPs and secondary care; the wish to expedite diagnosis, particularly in patients presenting at a later stage; the lack of clear ‘appropriateness’ guidelines, or other factors, drive GPs to send patients directly to emergency departments.