The nature and frequency of abdominal symptoms in cancer patients and their associations with time to help-seeking: evidence from a national audit of cancer diagnosis

Abstract Background Raising awareness of possible cancer symptoms is important for timely help-seeking; recent campaigns have focused on symptom groups (such as abdominal symptoms) rather than individual alarm symptoms associated with particular cancer sites. The evidence base supporting such initiatives is still emerging however; understanding the frequency and nature of presenting abdominal symptoms among cancer patients could inform the design and evaluation of public health awareness campaigns. Methods We examined eight presenting abdominal symptoms (abdominal pain, change in bowel habit, bloating/distension, dyspepsia, rectal bleeding, dysphagia, reflux and nausea/vomiting) among 15 956 patients subsequently diagnosed with cancer in England. We investigated the cancer site case-mix and variation in the patient interval (symptom-onset-to-presentation) by abdominal symptom. Results Almost a quarter (23%) of cancer patients presented with abdominal symptoms before being diagnosed with one of 27 common and rarer cancers. The patient interval varied substantially by abdominal symptom: median (IQR) intervals ranged from 7 (0–28) days for abdominal pain to 30 (4–73) days for dysphagia. This variation persisted after adjusting for age, sex and ethnicity (P < 0.001). Conclusions Abdominal symptoms are common at presentation among cancer patients, while time to presentation varies by symptom. The need for awareness campaigns may be greater for symptoms associated with longer intervals to help-seeking.


Introduction
Diagnosing cancer early in symptomatic patients is a prominent feature of contemporary cancer control strategies. 1,2 A range of pioneering studies during the last decade have established associations between the knowledge ('awareness') of likely symptoms of cancer among the general public and timely presentation, diagnosis, and outcomes. [3][4][5][6] Public health agencies have consequently implemented educational interventions aimed at raising awareness of cancer symptoms in order to promote timely presentation. [7][8][9] However, the evidence base supporting the design of such interventions is still emerging.
Previous symptom awareness campaigns have tended to take a cancer-based approach, by targeting 'red-flag' or 'alarm' symptoms explicitly associated with specific cancers, such as 'blood in poo' and colorectal cancer. [10][11][12] There is however growing interest in targeting symptoms relating to a body area or system, partly as this provides an opportunity to promote the earlier presentation of rarer and less common cancers. In England, an abdominal symptoms campaign was recently piloted at regional level, focusing on a range of symptoms (diarrhoea, bloating, abdominal discomfort, constipation, nausea, and blood in poo). 13 Examining the length of the patient interval (time from symptom onset to presentation) associated with different abdominal symptoms can contribute to the design of future campaigns. Awareness campaigns about possible cancer symptoms aim to shorten the patient interval by encouraging timely symptom appraisal and help-seeking. 14 Therefore, symptomspecific patient intervals may be interpreted as measures of relative need for such interventions. 15 Alongside considerations of other important factors such as the predictive value of a symptom for cancer, and the prevalence of different symptoms in the general population, such evidence can support how the content of awareness campaigns could prioritize certain symptoms over others.
Further, estimating the impact of a symptom awareness campaign has been shown to be challenging due to the diffuse and broad-reaching nature of campaigns; such difficulties are likely to be exacerbated by symptom-based approaches that target more than one cancer site. 7,16 Evidence regarding the anticipated cancer site case-mix of a particular symptom could help guide the direction of evaluation strategies, though such evidence is generally lacking. 17 We therefore examined the frequency of abdominal symptoms at presentation in a representative population of incident cancer patients; described the range of cancers associated with abdominal symptoms in an incident cohort; and examined variation in the length of the patient interval by presenting abdominal symptom.

Data source
We used data from the first English cancer audit (National Audit of Cancer Diagnosis in Primary Care) 2009-10, details of which have been described previously. 18 Briefly, participating clinicians collected information on the diagnostic process for incident cancer patients in~14% of all general practices in England, excluding screen-detected cases. The audited cancer patient population was representative of incident cancer patients in England during the same period, while the characteristics of participating practices were found to be comparable to non-participating practices. 18,19 Patient population We analysed data from cancer patients with complete and valid information on age group (among patients aged 15 years or older), sex, presenting symptoms and cancer site (see Supplementary Fig. S1 for flow chart of sample derivation). Individuals diagnosed incidentally and those with cancer sites categorized as 'No information' and 'Unknown Primary' were excluded from the analysis. Among the 3661 cancer patients with one or more abdominal symptoms, 2936 (80%) had complete information on the patient interval (see Supplementary Table S2 for the proportion of missing values by individual symptom). Overall, the strongest predictor of missing interval or pre-referral consultation data was first presentation to a healthcare facility other than the patient's own general practice, without evidence for substantive differences by socio-demographic characteristic (data not shown).

Variables of interest
General practitioners participating in the audit provided free text answers to the question 'what was the main presenting symptom?' for each patient, based on information in their primary care records. As described previously, 20 we coded symptom constructs following principles of natural language processing (NLP), without prior definitions or restrictions regarding cancer-symptom associations. 21 If multiple symptoms were mentioned, they were assumed to be synchronous. Symptoms were initially assigned by MMK, and cross-validated by GL and GPR, an approach also used previously. 20 Based on the abdominal symptoms described by the 2015 National Institute for Health and Care Excellence (NICE) guidelines for suspected cancer, we selected a total of 18 symptom constructs (see Supplementary Table S2) which were further aggregated into eight abdominal symptom groups: (non-acute) abdominal pain, bloating or distension, change in bowel habit, dysphagia, dyspepsia, nausea or vomiting, rectal bleeding and reflux. 22 The patient interval was defined as the number of days between symptom onset and the first presentation to primary care, in line with the Aarhus Statement. 23

Statistical analysis
The frequency (and associated exact confidence intervals) of abdominal symptoms in the studied population of cancer patients were estimated. We then described the cancer site case-mix of abdominal symptoms, namely the range and relative frequencies (proportions) of different cancer sites subsequently diagnosed among cancer patients presenting with abdominal symptoms.
Subsequently, we examined variation in the patient interval by abdominal symptom. As public awareness campaigns target individual symptoms rather than symptom combinations, these analyses were restricted to the majority of cancer patients with a single recorded presenting abdominal symptom (n = 2253, 62% of all patients reporting an abdominal symptom)though we examined common abdominal symptom combinations in supplementary analyses. Firstly, the mean, median, interquartile range and 90th centiles of the patient interval were estimated for each abdominal symptom along with 95% confidence intervals using a bootstrap approach with 1000 replications. Kruskal-Wallis tests were used to test variation in median interval length by abdominal symptom. The proportion of patients with each symptom that experienced a patient interval of 60 days or longer was also calculated to help to further contextualize the findings.
We then used generalized linear models (GLM) to examine the association between abdominal symptoms and the patient interval adjusted for age group (parameterized as <50 years, 50-69 years, 70+ years), ethnicity (white, non-white) and sex (men, women) given prior evidence supporting their associations with diagnostic timeliness. 24 To account for skewed outcome data, a log link function was used (which allows the covariates to be modelled on a linear additive scale, aiding interpretation), and significance testing was based on bootstrapping (1000 replications). Variation in interval length was examined using joint Wald tests, with statistical significance at the 5% level. All analyses were conducted using STATA SE v 13.1 (StataCorp, College Station, TX, USA).

Supplementary analyses
We conducted supplementary analyses examining the frequency of 12 most common abdominal symptom combinations, and their associated distributions of the observed patient interval in the same way as described above.

Frequency of presenting abdominal symptoms in cancer patients
Of a total of 15 956 patients with cancer, 3661 (23%) presented with one or more abdominal symptoms. Abdominal pain was the most common abdominal symptom across the entire cohort of cancer patients (8%), followed by change in bowel habit (6%), and rectal bleeding (5%) ( Table 1).

Cancer site case-mix of abdominal symptoms in cancer patients
Among the 3661 cancer patients who presented with abdominal symptoms, the majority (89%, 3244/3661) were diagnosed with solid cancers of abdominal or adjacent organs (Fig. 1). The most commonly diagnosed cancer site was colorectal cancer (47%), followed by oesophageal (13%), ovarian (7%) and pancreatic (6%) cancers (Table 2 and Fig. 1). A further 14 cancer sites were represented among the remainder of patients, including solid tumours of non-abdominal (and non-adjacent) organs (8%) and haematological cancers (4%).
We also considered the relative importance of abdominal symptoms for each cancer site by calculating the proportion of patients with a given cancer who had presented with one or more abdominal symptoms. Unsurprisingly, over twofifths (41%) of cancer patients diagnosed with an abdominal cancer had presented with abdominal symptoms, although this ranged from 84% of patients later diagnosed with oesophageal cancer to 5% of patients later diagnosed with prostate cancer (see Table 2 for full breakdown). Patients with cancers arising outside the abdominal region were much less likely to report abdominal symptoms (4%, n = 279). In contrast, patients diagnosed with haematological cancers were relatively more likely to report abdominal symptoms at presentation (11%, n = 138), almost two-thirds of those being patients with lymphoma (Table 2).

Patient interval by presenting abdominal symptom
Among cancer patients with a single presenting abdominal symptom (n = 2253), there was strong evidence for variation in the patient interval (symptom-onset-to-presentation) by symptom (P < 0.001, Fig. 2 and Supplementary Table S3). Patients presenting with change in bowel habit or dysphagia had the longest patient intervals: one in two patients with either of these symptoms waited at least a month before presentation, while a quarter waited 2 months or longer (median (IQR) patient interval: 30(4-73) days for change in  Table S4).
In supplementary analyses, we considered the 12 most common categories of single or combinations of presenting symptoms, including 3438 patients (94% of patients reporting one or more abdominal symptom). Results were largely comparable to the main analyses finding in respect of associations with the patient interval (see Supplementary Tables  S5 and S6).

Discussion
Main findings of this study Almost one in four cancer patients presented with abdominal symptoms before diagnosis. The majority of cancer patients who presented with abdominal symptoms were subsequently diagnosed with a range of common and rarer cancers of abdominal or adjacent organs, but a proportion of patients had tumours of other solid organ tumours, or haematological malignancies. The median patient interval ranged from 7 days for abdominal pain to 30 days for dysphagia. The observed differences in interval length by abdominal symptom remained when adjusted for age, sex and ethnicity.

What is already known on this topic
In our study, colorectal, oesophageal, ovarian and pancreatic cancers accounted for the majority of cancer patients that presented with one or more abdominal symptoms, consistent with previous evidence. 25,26 However, we also found large proportions of patients diagnosed with rarer cancers such as stomach (65%), small intestinal (69%) and gallbladder cancers (51%) presenting with abdominal symptoms.
Comparable evidence on the association between the patient interval and abdominal symptoms is limited to two English studies on colorectal and pancreatic cancers, respectively. 26,27 Rectal bleeding and dyspepsia-like symptoms were associated with shorter time to presentation compared with other studied symptoms, in line with our findings regarding these symptoms. 26

What this study adds
In order to improve the timeliness of diagnosis among cancer patients who present with symptoms, we need a better appreciation of the nature and frequency of presenting symptoms among these patients; currently, related epidemiological evidence is limited in quantity and breadth 17 . Consequently, our study adds substantially to the present evidence base, both regarding the burden of abdominal symptoms in incident cancer patients, and their associations with time to help-seeking. Abdominal symptoms appear to be common among incident cases of cancer, suggesting that symptom awareness campaigns focusing on abdominal symptoms could potentially expedite the diagnosis of a large range of both common and rarer cancers.
Previous analyses have shown large variation in the patient interval by cancer site. 28,29 Our findings suggest that this chiefly reflects variation in interval length of the most frequent symptoms of the different cancers. After considering symptom prevalence and predictive values of each symptom, variation in the length of the patient interval associated with different symptoms could help to identify particular symptoms for prioritization in campaigns. For example, we found that one in two cancer patients with dysphagia waited almost a month before presenting. As dysphagia is also an established 'alarm' symptom for cancer, this finding argues for its further targeting by future campaigns. 30 In contrast, cancer patients with abdominal pain presented after a median interval of 7 days, and given its high prevalence and low predictive value, there may be little to be gained by raising its awareness amongst the general population. 31,32 Previous evaluations have examined the increase in number of '2-week wait' referrals, the corresponding conversion rates to cancer cases, and diagnostic activity. 33 For campaigns targeting groups of symptoms, understanding the anticipated range of affected cancer sites will be crucial for accurate assessment of the campaign's impact.

Limitations of this study
The study design enabled analysis of data on both the presenting symptoms and associated patient intervals among a    [25][26][27]30,34 There are several limitations. Firstly, data on symptoms and the patient interval used in our study is reliant on the information on presenting symptoms and their duration being accurately and completely declared or elicited during consultation, and recorded in the patient's record. Nonetheless such approaches enable the profiling of large patient groups without potential concern about recall or survivorship bias. 29,35 A minority of patients with abdominal symptoms had missing outcome data regarding the patient interval, as noted in similar studies in this field. 26,27,36,37 We restricted our analyses to eight abdominal symptoms based on those recommended for urgent referral in national clinical guidelines. 22 This was a pragmatic decision that has face validity as symptom awareness campaigns are unlikely to include symptoms with a very low predictive value. We examined the patient interval among patients with a single abdominal symptom for ease of interpretation, again because campaign messages have thus far focused on single symptoms as opposed to synchronous symptom combinations. Further, in sensitivity analyses considering the most frequent symptom combination groups among nearly all cancer patients who presented with one or more abdominal symptom, we found concordant findings (see Supplementary  Table S5 and S6).
Our analysis focuses on the significance of abdominal symptoms among patients subsequently diagnosed with cancer, and provides insight into how awareness campaigns may be evaluated. Nevertheless, it is clear that abdominal symptoms in primary care may represent other important diseases, such as inflammatory bowel disease. 26,38 Coordinating our findings with evidence regarding the prevalence of abdominal symptoms among the general population, and the potential diagnostic experiences of patients that seek help for such symptoms beyond the cancer context may bring further insight. 39,40 We were unable to examine variation in patient interval by comorbidity status or deprivation: symptom appraisal and therefore the length of the patient interval, may be influenced by the presence of other conditions, and lower socioeconomic groups tend to experience lowest symptom knowledge and longer time to presentation. [41][42][43] However, such associations, if present, are unlikely to substantially confound the observed variation by abdominal symptom, which is the main focus of our study. Finally, while our findings provide insight into the associations between symptoms and timeliness of help-seeking before major population level campaigns were launched (in 2011), further examination of these associations between symptoms and timeliness of help-seeking in more recent cohorts will provide further insight. 33

Conclusions
In conclusion, almost a quarter of all patients with cancer initially present with an abdominal symptom, and their interval to presentation varies substantially by (abdominal) symptom type. The timeliness of presentation associated with individual symptoms could inform the design of campaigns, while the cancer site case-mix of a particular symptomatic presentation could be used to inform evaluation.

Supplementary data
Supplementary data are available at the Journal of Public Health online.