Quantifying the impact of pre-existing conditions on the stage of oesophagogastric cancer at diagnosis: a primary care cohort study using electronic medical records

Abstract Background Pre-existing conditions interfere with cancer diagnosis by offering diagnostic alternatives, competing for clinical attention or through patient surveillance. Objective To investigate associations between oesophagogastric cancer stage and pre-existing conditions. Methods Retrospective cohort study using Clinical Practice Research Datalink (CPRD) data, with English cancer registry linkage. Participants aged ≥40 years had consulted primary care in the year before their incident diagnosis of oesophagogastric cancer in 01/01/2010–31/12/2015. CPRD records pre-diagnosis were searched for codes denoting clinical features of oesophagogastric cancer and for pre-existing conditions, including those providing plausible diagnostic alternatives for those features. Logistic regression analysed associations between stage and multimorbidity (≥2 conditions; reference category: no multimorbidity) and having ‘diagnostic alternative(s)’, controlling for age, sex, deprivation and cancer site. Results Of 2444 participants provided, 695 (28%) were excluded for missing stage, leaving 1749 for analysis (1265/1749, 72.3% had advanced-stage disease). Multimorbidity was associated with stage [odds ratio 0.63, 95% confidence interval (CI) 0.47–0.85, P = 0.002], with moderate evidence of an interaction term with sex (1.76, 1.08–2.86, P = 0.024). There was no association between alternative explanations and stage (odds ratio 1.18, 95% CI 0.87–1.60, P = 0.278). Conclusions In men, multimorbidity is associated with a reduced chance of advanced-stage oesophagogastric cancer, to levels seen collectively for women.


Introduction
Diagnosing cancer early is a UK government priority (1). By 2028, the target is for 75% of UK cancers to be diagnosed at an early stage, leading to 55 000 more people annually surviving 5 years post-diagnosis (1). Recognizing cancer symptoms and prompt investigation are key to achieving this.
Around 50-60% of all cases are advanced-stage diagnoses (note incomplete staging data levels of 19% for oesophageal, 27% for stomach cancers), requiring more intense treatment, and with poorer outcomes (2,3). Diagnostic delay may contribute (6). Compared with rectal cancer, the odds of requiring three or more pre-referral primary-care consultations was higher for gastric [1.96, 95% confidence interval (CI) 1.65-2.34] and oesophageal (1.15, 0.98-1.36) cancers (7). This suggests the potential for earlier investigation in the disease process. A shorter diagnostic interval (time from first reported symptom to diagnosis) is associated with early-stage diagnosis of oesophageal cancer (8).
Patients with two or more pre-existing medical conditions (i.e. multimorbidity) may require complex management, with separate clinical pathways for each condition (9,10). This could expedite or delay diagnosis. Expedited diagnosis may occur in patients whose conditions require routine monitoring, by increasing opportunities for symptom reporting (the surveillance hypothesis) (11)(12)(13). In contrast, diagnosis may be delayed if cancer symptoms mimic those of the pre-existing condition (the alternative-explanations hypothesis) (13,14), or the primary-care consultations are dominated by condition management (the competing-demands hypothesis) (13).
Increased pre-existing condition count is associated with advanced-stage cancers of the ovary (15), larynx (16) and breast (12). In colorectal cancer, diagnostic intervals were longer by 32 days in patients with four or more conditions (17). Patients and clinicians may be more likely to normalize pre-existing symptoms, such as rectal bleeding, especially in patients with gastrointestinal conditions (18). One study investigating pre-existing multimorbidity in nine cancers found an increased chance of advanced-stage diagnosis in those patients with three or more conditions (19). Patients with upper gastrointestinal cancer (liver/gastric) had the highest cancerspecific index of multimorbidity (26%) of all nine cancers (19).
The impact of pre-existing conditions on the stage of oesophagogastric cancer at diagnosis is unknown. This study aimed to investigate whether oesophagogastric cancer stage is associated with: (i) pre-existing conditions that share symptoms of oesophagogastric cancer ('alternative-explanations' hypothesis) and (ii) multimorbidity ('competing-demands' or 'surveillance' hypotheses).

Methods
This retrospective cohort study of participants diagnosed with oesophagogastric cancer between 1 January 2010 and 31 December 2015 was set in primary care in England. Data sources were the UK's Clinical Practice Research Datalink (CPRD) GOLD database, with linkage to Public Health England's National Cancer Registration and Analysis Service (NCRAS, set 15) and Office for National Statistics (ONS) data. The CPRD holds anonymized longitudinal electronic records on symptoms, diagnoses, prescriptions and investigations of over 11.3 million patients from 674 UK general practices (20). Linked NCRAS data verified the date and type of diagnosis, and provided stage. Linked deprivation data (Townsend score) were obtained from ONS.

Inclusion and exclusion criteria
Participants were selected if they: Participants were excluded for missing NCRAS data on the best estimate of stage based on tumour size, nodal involvement and the presence of metastases (21).

Patient characteristics
Patient age and sex were provided by CPRD variables, assigning a birthday of 1 July. Patients were categorized from 1 (least deprived) to 5 (most deprived) using Townsend deprivation score data.

Stage and date at diagnosis
Stage at diagnosis was the outcome variable for regression analysis. Stage was classified as early (stage 1 or 2) or advanced (stage 3 or 4) using the NCRAS staging variable. Diagnosis date was provided by NCRAS.

Features of possible oesophagogastric cancer before diagnosis
Participants' CPRD records in the year before diagnosis were searched for codes for features of possible oesophagogastric cancer: dysphagia, dyspepsia, reflux, weight loss, upper abdominal pain, nausea, vomiting, haematemesis and anaemia (4). Relevant Read code lists were assembled using an established protocol (22). Participant-level variables described the first possible feature of cancer (hereafter called 'index feature'), with category '0' denoting asymptomatic status. The 'index date' was the date of the index feature, or the diagnosis date for asymptomatic participants.

Pre-existing conditions
The QOF conditions were: anxiety and/or depression; asthma; atrial fibrillation; chronic obstructive pulmonary disease; coronary heart disease; dementia; diabetes mellitus; epilepsy; heart failure; high blood pressure; hypothyroidism; learning disability; mental health disorders (bipolar disorder, schizophrenia and other psychoses); osteoporosis; peripheral arterial disease and rheumatoid arthritis.

Explanatory variables denoting multimorbidity and alternative explanations
Participant-level indicator variables identified participants with two or more pre-existing conditions (i.e. multimorbidity) before the index date. The numbers (percentages) of participants (by sex) with each pre-existing condition are reported.
Participant-level indicator variables denoting 'alternative explanations' identified participants with one or more pre-existing conditions that provided a plausible explanation for their index feature(s).

Data analysis
The cohort is summarized using descriptive statistics. The numbers with each comorbid condition and with alternative explanations are reported by sex, for all participants included in analysis or excluded for missing stage. Associations with sex were explored using the chi-square test.
Associations between stage at diagnosis (reference group: early stage) and both multimorbidity (reference group: no multimorbidity) and the presence of alternative explanations (reference group: no alternative explanations) were analysed using multi-level logistic regression. Interaction terms were sought on clinical grounds between sex and: (i) multimorbidity, and (ii) the presence of an alternative explanation; and (iii) between multimorbidity and age. The analyses accounted for the correlation among individuals within the same general practices (clusters) and adjusted for possible confounding by age at diagnosis, sex, deprivation and cancer site. Robust standard errors accounted for heteroscedasticity and the analysis was repeated with log(age) and with age-squared to check for non-linearity. Post-estimation diagnostics tested for model specification (linktest), goodness of fit (lfit) and collinearity between explanatory variables (collin, Collinearity Diagnostics, Philip B Ender, UCLA Office of Academic Computing).
The regression coefficients are presented. To ease interpretation, after running the final model, we used Stata's margins commands to estimate the predicted probability of having an advanced-stage diagnosis for the following groups of interest, holding other variables to their mean values: 1. Men and women with multimorbidity.

Men and women with alternative explanations.
Data analysis was conducted using Stata (version 16) (StataCorp, College Station, TX).

Missing data, bias and sensitivity analysis
Both CPRD and NCRAS have missing data. In line with convention, we interpreted the absence of a code for a clinical event as its nonoccurrence (14,20,24). We classified participants with no recorded features of cancer and/or no diagnostic codes for comorbid conditions as not presenting with alternative explanations.
To examine for bias, we conducted a logistic regression to compare participants with and without staging data with regard to the explanatory variables for our main model, including interaction terms between multimorbidity and age or sex.
In sensitivity analysis, we re-ran the final model, including participants with missing staging data assigned to advanced-stage diagnosis.

Power calculation
The complete-case sample of 1749 patients had >95% power to detect a change of 10 percentage points in the proportion diagnosed with advanced disease between those with and without multimorbidity (two-sided, alpha = 0.01). This is based on 60% of participants being multimorbid and 70% having advanced-stage disease (9,10).

Index features of cancer
Most included participants (1387/1749, 79.3%) had recorded features of possible oesophagogastric cancer before diagnosis (

Stage at diagnosis
Advanced-stage oesophagogastric cancer was diagnosed in 1265/1749 (72.3%) of included participants (909/1236, 73.5% for men; 356/513, 69.4% for women). The proportion with multimorbidity was higher in those with early-stage (71.1%) than with advanced-stage (62.6%) disease. The proportions with alternative explanations for their index features of cancer were similar in patients with early or with advanced-stage disease ( Table 2).
Having alternative explanations was not associated with stage (Table 4). Consequently, the probabilities of advanced-stage disease were similar in people with (0.74, 0.69-0.78) and without (0.72, 0.70-0.75) alternative explanations.
Post-estimation regression diagnostics suggested no problems in model specification, goodness of fit or collinearity between explanatory variables (results available from authors). Participants were aged ≥40 years and had attended their CPRD general practice at least once in the year before their incident diagnosis with oesophagogastric cancer in the period 1 January 2010 to 31 December 2015. Participants with features of possible cancer presented with any of dysphagia, dyspepsia and/or reflux, weight loss, upper abdominal pain, nausea, vomiting, haematemesis or anaemia in the year before diagnosis. Multimorbidity was defined as having at least two conditions diagnosed before either the first possible feature of cancer or cancer diagnosis (for participants who did not present with cancer features). Participants with alternative explanations had pre-existing conditions providing a plausible diagnostic alternative for their cancer feature.
In 'missing-is-advanced-stage' analysis, the coefficients were similar to those in the main model, apart from age (1.00, 95% CI 0.99-1.01, P = 0.683) (see Supplementary Material).

Summary
This is the first primary-care study to examine the association between pre-existing medical conditions and the stage of oesophagogastric cancer at diagnosis. We provide some evidence of differing effects of multimorbidity between the sexes. For men, being multimorbid reduces the chance of advanced-stage oesophagogastric cancer to levels seen for women regardless of their multimorbidity status, controlling for age, deprivation and cancer site. Having alternative explanations for features of possible oesophagogastric cancer is not associated with stage.

Strengths and limitations
The study analysed 1749 primary care patients in England, ensuring ample power. Our data source was the CPRD, which has NCRAS linkage, and is large and representative, enabling generalizability of our results (20). The demographics of our sample are consistent with the epidemiology of oesophagogastric cancer, in terms of age at diagnosis, sex and deprivation profiles, and unknown stage (2,3). A further strength is the primary-care setting, where general practitioners face the difficulty of recognizing which patients need referral or investigation for undiagnosed cancer.
Our main limitations relate to missing CPRD and NCRAS data. First, when using CPRD data, we depend on how GPs choose which clinical details to record and their recording method. Our use of QOF conditions likely minimizes any underestimation of multimorbidity, as QOF conditions are well-defined and coded recording is encouraged by linkage to practice payments (23). Test results have automated transmission, with have low levels of missingness (20). However, some symptoms may have been omitted, or recorded in the 'free-text' section-which is inaccessible to researchers. For vague symptoms especially, up to one-third of affected participants may have symptom records 'lost' in free text (25). Symptom underrecording may arise because of time-pressured consultations, cognitive factors influencing clinical assessment and diagnostic reasoning (26,27). We may have underestimated the numbers with alternative explanations, reducing our power to identify associations between alternative explanations and stage.
Our study is limited by having to exclude 695 (28.4%) of the sample with no recorded stage. Incomplete staging data for oesophagogastric cancer is a known issue (2,3). Stage was plausibly missing at random, conditional on increasing age and female sex. This is consistent with patients who are elderly or who present as an emergency (more likely in women) having less comprehensive CHD, coronary heart disease; CKD, chronic kidney disease; COPD, chronic obstructive pulmonary disease; IBD, inflammatory bowel disease; IBS, irritable bowel syndrome. Other conditions in men and women, respectively: epilepsy (n = 27, n = 7); dementia (n = 10, n = 10); Parkinson's (n = 4, n = 1); pancreatitis (n = 1, n = 1); oesophageal stricture (n = 2 men); learning disability (n = 4, n = 3).
P values are from chi-square tests. Gender differences for participants included in analysis highlighted in bold.
diagnostic and staging investigations (13,28,29). It is reassuring that the distributions of age and sex in our sample included in the analyses are very similar to national figures (2,3), suggesting that any bias was small. Furthermore, in sensitivity 'missing-is-advanced' analyses, the coefficients for multimorbidity, sex and their interaction were similar in size to those in the main model.

Comparison with previous literature
Previous research established associations between multimorbidity and advanced-stage diagnosis of ovarian, laryngeal, breast and colorectal cancers (12,(14)(15)(16). Findings related to gastric/liver cancers were inconclusive, although separate effects for gastric and liver cancers or for men and women were not investigated (19). Our findings are consistent with the surveillance hypothesis effect of multimorbidity in oesophagogastric cancer for men (7,13). Men consult primary care less frequently than women, after allowing for women's consultations for reproductive health care; however, the difference was not observed between men and women with comparable morbidities (30). We found no association between stage and having alternative explanations; therefore, our study does not elucidate why women are more likely than men to be diagnosed following an emergency presentation (2,3,5).
Our findings contrast with our previous study of bladder cancer (14). For that cancer, we found no association between count of pre-existing conditions and stage, but a strong association between alternative explanations for features of bladder cancer and advancedstage diagnosis (14). This difference may reflect the symptomatology of the two cancers: haematuria is the dominant symptom and highly predictive of bladder cancer (14,31), and is recommended for urgent investigation (4). In contrast, in our study, less than one-third of oesophageal cancers presented with the high-risk symptom of dysphagia, the majority presenting with low-risk features. It may be that the increased primary-care attendances trigger investigation of lowrisk symptoms, such as dyspepsia, though not necessarily with cancer as the condition being sought. There is a clear relationship between a primary-care practice's increased gastroscopy rates and improved oesophageal cancer outcomes, supporting this interpretation (32).

Implications for research and practice
This study is consistent with the surveillance hypothesis, at least in men. It has always been considered good primary-care practice to extend the consultation beyond the presenting complaint, by elucidating other concerns, offering health care advice and preventative care. These 'supplementary' aspects require time, which is not always available within a single consultation. This study cannot provide a definite mechanism for the findings, being observational, but suggests that extra patient contact may have a measureable benefit. Further planned research will explore possible mechanisms by investigating gender differences in the recognition of cancer symptoms, and assessment of cancer risk in multimorbidity.

Supplementary material
Supplementary material is available at Family Practice online.

Data availability
The anonymized patient data from this study are not available due to legal privacy restrictions enforced by the CPRD. Code lists and symptom libraries are available from the authors by request. Multimorbidity was defined as ≥2 pre-existing conditions. Participants with 'alternative explanations' had a pre-existing condition that provided a plausible diagnostic alternative for their presenting feature of oesophagogastric cancer (i.e. dysphagia, dyspepsia and/or reflux, weight loss, upper abdominal pain, nausea, vomiting, haematemesis or anaemia). Note: Odds ratios (ORs) >1 indicate increased odds of advanced stage at diagnosis and ORs <1 indicate reduced odds. An 'alternative explanation' is a condition that provides a plausible diagnostic alternative for the index cancer symptom (reference group: no alternative explanation). Multimorbidity is defined as two or more pre-existing conditions (reference group: not multimorbid). For sex, the reference group is men.