Emergency diagnosis of cancer and previous general practice consultations: insights from linked patient survey data

Background Emergency diagnosis of cancer is common and aetiologically complex. The proportion of emergency presenters who have consulted previously with relevant symptoms is uncertain. Aim To examine how many patients with cancer, who were diagnosed as emergencies, have had previous primary care consultations with relevant symptoms; and among those, to examine how many had multiple consultations. Design and setting Secondary analysis of patient survey data from the 2010 English Cancer Patient Experience Survey (CPES), previously linked to population-based data on diagnostic route. Method For emergency presenters with 18 different cancers, associations were examined for two outcomes (prior GP consultation status; and ‘three or more consultations’ among prior consultees) using logistic regression. Results Among 4647 emergency presenters, 1349 (29%) reported no prior consultations, being more common in males (32% versus 25% in females, P<0.001), older (44% in ≥85 versus 30% in 65–74-year-olds, P<0.001), and the most deprived (35% versus 25% least deprived, P = 0.001) patients; and highest/lowest for patients with brain cancer (46%) and mesothelioma (13%), respectively (P<0.001 for overall variation by cancer site). Among 3298 emergency presenters with prior consultations, 1356 (41%) had three or more consultations, which were more likely in females (P<0.001), younger (P<0.001), and non-white patients (P = 0.017) and those with multiple myeloma, and least likely for patients with leukaemia (P<0.001). Conclusion Contrary to suggestions that emergency presentations represent missed diagnoses, about one-third of emergency presenters (particularly those in older and more deprived groups) have no prior GP consultations. Furthermore, only about one-third report multiple (three or more) consultations, which are more likely in ‘harder-to-suspect’ groups.


INTRODUCTION
Evidence from several countries documents that many patients with cancer are diagnosed through an emergency presentation (hereafter, the term 'emergency presenters' is used for these patients). [1][2][3][4] As emergency presentations are associated with poorer survival and worse patient experience, reducing their frequency is desirable, but how to achieve such reductions is uncertain. [5][6][7][8] Complex aetiologies, reflecting different disease (tumour), patient, and healthcare factors, often in combination, are likely to be implicated. [9][10][11] Prior evidence indicates that older patients and patients who are more socioeconomically deprived are at substantially higher risk of diagnosis of cancer as an emergency. 1,3,4 An ecological study indicates a degree of variation by certain general practice characteristics, but overall there is very limited evidence about whether and how diagnoses of cancer through an emergency presentation can be avoided. 11,12 A key consideration is whether emergency presentations were preceded by patient contact with the healthcare system. Some emergency presenters would have had no such prior contact. This can occur either when dramatic (life-threatening) clinical presentations are preceded by minimal or short-lived symptoms, making emergency presentations practically unavoidable (indeed representing optimal care); or when patients have experienced, often mild, symptoms for a long time but have not sought medical help until sudden changes in the nature or severity of their symptoms prompted them to do so in an emergency context. 9,10 Practical (for example, access), cognitive (for example, symptom awareness), or emotional (for example, fear of cancer diagnosis) factors may all act as barriers to help seeking.
Other emergency presenters would have sought medical help for their symptoms previously (often in primary care), but either referrals or investigations were not initiated; or a diagnostic process was instigated but sudden subsequent changes in symptom nature or severity led to an emergency presentation. 9,10 Additionally, emergency presentations may result from contact with a GP who then directly refers the patient as an emergency to the hospital. 13 As also highlighted by the findings of a recent in-depth review of the diagnosis of cancer as an emergency, evidence on Research Abstract Background Emergency diagnosis of cancer is common and aetiologically complex. The proportion of emergency presenters who have consulted previously with relevant symptoms is uncertain.

Aim
To examine how many patients with cancer, who were diagnosed as emergencies, have had previous primary care consultations with relevant symptoms; and among those, to examine how many had multiple consultations.

Design and setting
Secondary analysis of patient survey data from the 2010 English Cancer Patient Experience Survey (CPES), previously linked to populationbased data on diagnostic route.

Method
For emergency presenters with 18 different cancers, associations were examined for two outcomes (prior GP consultation status; and 'three or more consultations' among prior consultees) using logistic regression.
the proportion of emergency presenters with and without previous consultations is very limited. 11 This study aimed to examine variation (by patient characteristic and cancer site) in the presence and number of prior consultations with a GP among emergency presenters, to provide insights into aetiological mechanisms responsible for such presentations.

METHOD Data
A secondary analysis was performed of data from the 2010 English Cancer Patient Experience Survey (CPES), previously linked to information about diagnostic 'route' (see below) as part of a national report on cancer patient experience. 8,14 The patient survey was commissioned by the UK Department of Health and carried out by Quality Health, a specialist survey provider. 12 The sampling frame included patients aged ≥16 years with a cancer diagnosis, seen as inpatients or day cases in English NHS hospitals during January-March 2010. After vital status checks, patients were sent the survey questionnaire by post a few weeks after their treatment, with up to two reminders for non-responders.
Analyses were restricted to responders who were emergency presenters, based on data linkage with the Routes to Diagnosis dataset, carried out previously by the Public Health England (former) National Cancer Intelligence Network to support public reporting of data on cancer patient experience (Appendix 1). 8 Diagnostic 'routes' represent different care pathways to cancer diagnosis. 1 The emergency presentation route denotes new diagnosis of cancer following accident and emergency department attendance, emergency hospital admission, emergency between-hospital transfer, or emergency GP referral (operational definitions have been detailed previously). 1 Outcomes. Variation was examined in respect of two outcomes: First, prior consultation status, that is whether emergency presenters had no prior GP consultations; and, second, 'multiple' consultation status, that is whether emergency presenters who had seen a GP had three or more consultations. Both outcomes were defined using information from the first survey question: 'Before you were told you needed to go to hospital about cancer, how many times did you see your GP (family doctor) about the health problem caused by cancer?'. Informative response options included 'None -I did not see my GP before going to hospital' (denoting 'no prior GP consultation' in the analysis) and 'Once', 'Twice', 'Three or four times', and 'Five or more times' (the latter two categories used to denote 'three or more' consultations among patients reporting at least one consultation).
The latter outcome was dichotomised (three or more versus one to two consultations), consistent with public reporting conventions for this survey, and because some repeat (second) appointments are generated by the need to review the results of investigations ordered at an initial consultation. The number of prediagnostic consultations is associated with the primary care interval (number of days between presentation in primary care and referral), with primary care intervals of 34, 47, and 97 days for the average patient with three, four, and 'five or more' consultations, respectively. 15 Exposures. These included patients' age group, sex, deprivation group, and cancer site using the International Classification of Diseases 10th edition (based on hospital records information included in the CPES dataset); and self-reported ethnic group (derived from survey responses) using a binary white/non-white variable because of the small numbers in minority ethnic

How this fits in
Diagnosis of cancer as an emergency has been considered to represent 'a failure of primary care'. Evidence to support such assertions is, however, limited. A notable minority of all emergency presenters have no prior contact with primary care, particularly males, older, and more deprived patients, and those with brain cancer. Among emergency presenters who have seen a GP, multiple consultations occur in a minority and are driven by diagnostic difficulty.

Box 1. Prior hypotheses
It was a priori hypothesised that the symptom signature of different cancers would be the main driver of variation by cancer. Specifically, it was hypothesised that in respect of symptom signatures: • Emergency presenters with cancers that often present acutely with symptoms deemed a clinical emergency (for example, seizures in patients with brain cancer) would probably have low proportions of prior consultations. • Emergency presenters with 'harder-to-suspect' cancers (for example, multiple myeloma) would probably have consulted multiple times.
Additionally, it was hypothesised that emergency presenters with leukaemia (where emergency presentations may result from abnormal findings in full blood count tests carried out in primary care), who were prior consultees, would probably not have consulted multiple times.
groups. Deprivation groups were defined using quintiles of the English Index of Multiple Deprivation 2007 score based on patients' postcodes.
Sample derivation. Of an initial total of 67 713 responders, diagnostic route information was available for 56 363 (83%). For practical reasons, analyses were restricted to emergency presenters with one of the 18 most common cancers. The final analysis sample included 4647 emergency presenters with complete information for all covariates (Appendix 2).
Analysis. Prior hypotheses regarding variation in both outcomes are shown in Box 1. Logistic regression models were used to estimate crude and adjusted odds ratios (ORs). The adjusted models included cancer site and sociodemographic variables. The survey sample is drawn from patients with cancer with recent hospital treatment rather than incident cases. Variations in treatment modality by cancer site distort the sample's composition compared with incident cases. 16 Further sample distortions occur because of post-sampling mortality and survey non-response. 5 Therefore, there are likely compositional differences between the analysis sample and the population of prior interest (incident emergency presenters). To account for this, when estimating overall percentages post-stratification weights were produced, derived using the complete (as opposed to linked) 'Routes-to-diagnosis' 2006-2010 dataset from a previous study (Appendix 3). 4 As all variables used in the derivation of weights were included in the adjusted models, weighting was not used in the regression analysis. Assuming the models are correctly specified, estimated OR should be unbiased and standard errors would be smaller than if weights had been included. As a sensitivity analysis, the regression analysis was repeated including weights.

RESULTS
The characteristics of the 4647 emergency presenters included in the analysis are shown in Table 1. Older and more deprived emergency presenters were underrepresented in the analysis sample, with only small differences by sex. Patients with haematological cancers (multiple myeloma, Hodgkin and non-Hodgkin lymphomas, and leukaemia) were over-represented and those with lung and pancreatic cancer under-represented (Appendix 3).

Outcome 1: emergency presentation without prior consultations
Among the 4647 emergency presenters, 1349 (29%) reported no prior consultation with a GP (Table 1). Using post-stratification weights (to account for distortions caused by sampling, non-response, and postsampling mortality), 34% of all emergency presenters with the studied cancers reported no previous consultations.
The proportion of emergency presenters without prior consultation varied greatly by cancer site, as indicated by sevenfold difference in adjusted odds (P<0.001). Specifically, emergency presenters with brain, renal, endometrial, or breast cancer were most likely to report no prior consultations (adjusted OR of 2.08, 1.71, 1.54, and 1.42, respectively, using colon cancer as reference); while those with mesothelioma, Hodgkin lymphoma, ovarian, or pancreatic cancer were least likely (adjusted OR of 0.30, 0.35, 0.41, and 0.45, respectively, Table 1, Figure 2). Outcome 2: three or more previous consultations (among emergency presenters with at least one consultation) Among the 3298 emergency presenters with at least one prior consultation, 1356 (41%) had three or more consultations (Table 2). Using post-stratification weights, 35% of all emergency presenters who report that they have consulted previously with a GP would have done so three or more times. This is equivalent to 23% (weighted percentage) of all emergency presenters (independently of prior consultation status). Three or more consultations were more common in females (45% versus 38% in males), younger patients (48% in 16-24-yearolds versus 39% in 65-74-year-olds), and ethnic minority patients (52% versus 41% for white patients), with little variation by deprivation group. Multivariable logistic regression provided concordant findings (Table 2, Figure 3).
There was large variation by cancer, as indicated by fourfold variation in adjusted odds (P<0.001), with three or more consultations being most likely in emergency presenters with multiple myeloma (OR 1.81) and least likely in those with leukaemia (OR 0.41) (Table 2, Figure 4).
Sensitivity analysis. Including poststratification weights in multivariable regression models provided broadly consistent findings (Appendix 4). Although the OR for some cancers changed considerably, these changes were small relative to the width of associated confidence intervals. This highlights that the exact ordering of cancers should not be over-interpreted, rather taken as a broad indication of its position. As expected, weighting increased the standard errors, resulting in some of the smaller variations no longer being significant.

DISCUSSION Summary
About one-third of emergency presenters reported no prior consultation with a GP. Older and more deprived patients, and those with brain cancer, were more likely not to have consulted previously. Of those who had previously seen a GP, younger and ethnic minority patients, and those subsequently diagnosed with 'harder-to-suspect' cancers were more likely to have consulted multiple times.

Strengths and limitations
Patient reports of consultation history were used, which directly relate to perceptions of their own healthcare experience and have high face validity. Patients might overestimate pre-referral consultations compared with information in medical records, but recall inaccuracies would likely introduce random error and could not explain the large observed variations. 17,18 Many different cancer sites were examined, enabling stronger inferences to be made on disease factors contributing to emergency presentations. The sample included emergency presenters with recent hospital treatment, not all emergency presenters among incident cases; however, the weighting of overall proportions took into account compositional differences between the two populations. 14 Weighting assumed that emergency presenters who died early or did not respond were 'missing at random' conditional on their age, sex, deprivation, and cancer site. This assumption may not be true, as emergency presenters in the sample may have different consultation patterns than otherwise similar (in terms of age, sex, deprivation, and cancer site) emergency presenters among incident cases. The potential for such bias is limited, however, as only 6% of all patients died between treatment and survey mail out, and the survey had a relatively high response rate (67%). 14 Previous related research has shown the effect of sociodemographic factors varying by cancer site, but power considerations meant that such interactions could not be considered in this study. 4,15 While the present findings indicate that two-thirds of emergency presenters had at least one prior primary care consultation, previous analysis indicates that GPs are directly involved (that is, through direct emergency referral to hospital services) in one-third of all emergency presentations. 11 Assuming that emergency presenters who were referred to hospital as emergencies by their GP would consider such involvement to be a prior consultation, the combined interpretation of these figures would indicate that at least one-third of all emergency presenters have had prior contact with GPs but were not referred as emergencies. With the present data, it is impossible to establish

Ethical approval
The research used entirely anonymous previously linked data, requiring no ethics approval.

Provenance
Freely submitted; externally peer reviewed.

Competing interests
The authors have declared no competing interests.

Open access
This article is Open Access: CC BY 4.0 license (http://creativecommons.org/ licenses/by/4.0/). the National Cancer Intelligence Network (currently the National Cancer Registration and Analysis Service), without whom there would be no cancer registration or Routes to Diagnosis data.

Discuss this article
Contribute and read comments about this article: bjgp.org/letters the proportion of emergency presentations generated by direct emergency GP referrals.

Comparison with existing literature
Prior evidence indicates that few emergency presenters had no prior primary care consultations, but is restricted to only three cancers (ovarian, lung, and colorectal) and dominated by medical record studies not examining whether the consultations were on relevant symptoms, and whether they were single or multiple. 2,[19][20][21][22][23][24] Males, older patients, and the most deprived emergency presenters were more likely to report no prior consultations, possibly reflecting a higher concentration of practical, cognitive, or emotional barriers to presentation in these patient groups. [25][26][27][28] It has been further suggested that dementia syndromes may be implicated in emergency presentation in some older patients. 29 Some deprived patients may prefer using emergency services for regular care, which may also partly explain the findings. 30 It has long been hypothesised that patients with brain cancer have a relatively high proportion of emergency presentations (>60%) because they often first present with seizures, without prodromal non-acute symptoms. The present findings support this hypothesis, which remains otherwise poorly studied.
It is important to note that the number of pre-referral consultations before a cancer diagnosis cannot be assumed to be a measure of diagnostic quality per se for individual patients. In some patients multiple consultations may represent appropriate care, including, for example, if there is a patient preference for non-referral during earlier consultations, or because of the need to review results of investigations ordered. However, in some patients (particularly when an emergency presentation has ensued), multiple consultations may represent potential for missed diagnostic opportunities. Where such judgements are possible, they can only be made after thorough case note review and the consideration of the presenting symptoms. 31

Implications for research and policy
Contrary to suggestions that emergency presentations represent 'failures of primary care', the present findings suggest that many emergency presenters have no prior contact with primary care and emphasise the potential influence of psychosocial patient factors. 32 The fact that males, older patients, and more deprived emergency presenters are more likely not to have consulted provides opportunities for targeting of general public health education interventions about cancer symptoms. 24 One-quarter of all emergency presenters have three or more prior GP consultations. As hypothesised, among emergency presenters with at least one prior consultation, those with 'harder-to-suspect' cancers (such as multiple myeloma and lung cancer) or characteristics associated with greater diagnostic difficulty (for example young age) were more likely to have had multiple consultations. 15 In these patients, emergency presentations seem to reflect the challenges of suspecting cancer when symptoms are vague, and/or the baseline risk of cancer is low.
Among emergency presenters with breast and endometrial cancer (who had consulted previously), 21% and 33% had three or more consultations, respectively. These proportions are appreciably higher compared with those observed in a typical female with either cancer (7% and 17%, respectively). 15 This suggests that emergency presenters with 'easy-to-suspect' cancers (such as breast and endometrial cancer) may tend to have atypical symptomatic presentations.
Consistent with the study hypotheses, patients with leukaemia had the lowest odds of three or more consultations among prior consultee emergency presenters, possibly reflecting direct emergency referrals by GPs after abnormal full blood count tests.
Generally, interventions aimed at improving diagnostic timeliness after presentation may also reduce emergency presentations. 33 Such interventions may include greater use of clinical decisionsupport tools, development of point-ofcare tests, and accelerated diagnostic care pathways to specialist assessment and imaging or endoscopic investigations. 34,35 In conclusion, against suggestions that emergency presentations represent missed diagnoses, about one-third of emergency presenters have no prior GP consultations, and only about one-third experience multiple consultations. Both disease (for example cancer site and symptom signature) and patient factors (for example, lower levels of symptom awareness in older and more deprived patients) are likely to be implicated in emergency presentations without prior consultation. The findings can guide future research and policies, focusing on public health education campaigns (for example, aimed at changing patient awareness, beliefs and behaviour, particularly in higher risk groups), or healthcare (diagnostic safety) interventions, variably targeting different patient groups and symptomatic presentations.   The stratification weights are standardised so that values >1 mean the emergency presentation of a type of cancer is under-represented in the English Cancer Patient Experience Survey (CPES) (and vice versa for values <1). These weights can be conceptualised as indicating the number of incident patients with cancer who present as emergencies in the population represented by each emergency presenter in the analysis sample. a These are the CPES 2010 responders who were diagnosed through emergency presentation. b The weights are calculated as follows: