Population and study setting

This multicenter, cross-sectional study involved patients in five regions of Spain (Aragón, Balearic Islands, Barcelona, Galicia and Valencia). Consecutive patients diagnosed with CRC (International Disease Classification 153–154) and registered with a GP were recruited through the pathology services of nine public hospitals between September 2006 and September 2008. Patients with prevalent or recurrent CRC or multiple tumors and patients diagnosed in private hospitals were excluded. Patients were contacted by their oncologist during the inpatient stage or during an outpatient oncology visit. Individual specialists invited these patients to participate in this study. All participants provided written informed consent. The methods used in this study have been published elsewhere [19–20].

Data collection procedures

Data were obtained by specifically trained GPs and nurses from patient interviews, together with reviews of primary care and hospital records. Data obtained from patient interviews included socio-demographic factors, such as age, sex, marital status, and level of education, history of cancer in family members or acquaintances, and initial symptom/s: Each patient was asked how long he/she had been feeling unwell. Symptoms spontaneously mentioned by the patient were considered as the initial symptoms for that patient and the date was recorded. If the patient could remember the exact day, then that date was recorded. If the patient could not accurately remember the date of onset then the approximate date was recorded; for example, if a patient reported symptom onset two months earlier, the date recorded was two months before the date of the interview. After recording voluntary data on disease onset, the interviewer asked each patient if he/she presented with any of the other symptoms on a 22 symptoms checklist. To assess perception of symptom seriousness, the patient was asked if he/she considered the initial symptoms very serious, quite serious, not serious, or other. To assess help-seeking action, the patient was asked what he/she did after the onset of first symptoms: visit a doctor, wait for symptoms to clear, or other. The type of doctor contacted and confidence in their GP (0–10, with 10 considered maximum confidence) were also recorded. The patients were interviewed after a median of 47 days after the diagnosis (IQR = 88 days).

Hospital records were reviewed after the interview. After determining the date of diagnosis from the pathology report, the data manager reviewed the records to determine the first patient contact related to CRC symptoms. Date of diagnosis was determined from the first pathology report. The first hospital service that evaluated the patient was dichotomized to presentation at the emergency department (EP) or to outpatient services (OS) (surgery, gastroenterology, internal medicine or others).

Primary care records were reviewed after the interview and the review of hospital records in order to have the precise date of diagnosis. Primary care records were reviewed for 2 years prior to the date of diagnosis to determine the first GP contact for CRC symptoms. If patients did not contact their GPs for CRC symptoms, no other data were collected. If patients did contact their GPs, then the data recorded included the number of subsequent visits for CRC symptoms to GPs prior to diagnosis as well as the number of primary care visits to a GP or nurse during the 12 months before the diagnosis, and any suspicion of diagnosis registered in a patient’s clinical record. In addition, the Charlson Comorbidity Index at diagnosis was recorded from both hospital and primary care clinical records.

Symptom duration was divided into several intervals and calculated in days. 1) The patient interval was defined as the number of days from the onset of first CRC symptoms to first consultation with a physician; 2) the health services interval was defined as the time from first contact with a physician to diagnosis; 3) the diagnosis interval was defined as the time from the onset of first CRC symptoms to the date of diagnosis; and 4) the total interval was defined as the time from first presentation with CRC symptoms presentation to date of treatment. Patient interval and total interval were calculated according to the Aarhus Statement [21].

Statistical analysis

Data are presented as numbers and percentages or as medians and interquartile ranges (IRs). Data on patients with colon and rectal cancer were examined separately throughout. Independent variables associated with the crude proportions of patients diagnosed through the EP or other hospital services were assessed by the Chi-Square test, whereas the association between differences in symptom duration and emergency presentation were analyzed using the Mann Whitney U-test. A P-value <0.05 was considered statistically significant.

Factors independently predictive of emergency presentation or of initial consultation with a GP or other doctor were analyzed by adjusted multivariate logistic regression analysis, as was GP performance among patients who consulted a GP. Any factor found by univariate analysis to be significant at P≤0.10 was included in the multivariate analysis. Interactions between factors included in the final models were examined. Sensitivity analyses of the models assessing emergency presentation and initial consultation were performed, without excluding the variables ‘first contact with health services’ and ‘hospital service of first referral’, respectively, as they could confound the associations of the other independent variables. All statistical analyses were performed using the Statistical Package for Social Sciences (SPSS version 23.0).

This study was approved by the Primary Health Care Committee of each health district and by the Ethical and Clinical Research Committee of each participating region (Comitè d'Ètica de la Investigació de les Illes Balears; Comité Ético de Investigación Clínica (CEIC) del Hospital Clínico Universitario de Valencia; Comité Ético de Investigación Clínica de Galicia; Comitè Ètic d'Investigació Clínica del Parc de Salut Mar; Comité Ético de Investigación Clínica de Aragón).

Strengths and limitations

One of the strengths of the study was our inclusion of important patient-reported data, providing insight into patient perception of symptoms and help-seeking behavior after symptom presentation. In addition, our inclusion of medical records compiled by GPs enabled our inclusion of important data reported by patients and considered relevant by their doctors.

Most of the limitations of this study have been described elsewhere [20–21]. First, there is no universally accepted definition of an emergency diagnosis of cancer. This constitutes the cornerstone for obtaining robust and comparable measures of different routes of diagnosis. As in most studies using routine data, some of these data were incomplete, limiting our ability to examine certain key characteristics and the confidence of our conclusions. Our findings may have been affected by missing data for some variables, mainly those from GP clinical records. Inclusion of patients diagnosed by screening or incidentally may also have biased our results. We found that a small proportion of these patients contacted a hospital through emergency services, perhaps because a GP or gastroenterologist referred these patients to emergencies based on the results of endoscopy or after an incidental detection of cancer in order to overcome waiting lists. Patient and GP recall of first symptoms did not always align, particularly for vague symptoms [22], and there were differences in recalling the date of first symptom presentation. Furthermore, the percentage of patients reported visiting a GP after symptom onset was higher than that reported in their primary care records, indicating that patients may have overestimated their contact with a GP or that some GPs may not have recorded vague symptoms.

Comparison with other studies and interpretation of our findings

The percentage of CRC patients who initially presented to emergency services was much higher in our study than in studies performed in other European and North America countries, which have been reported to range from 17–26% [6; 23–26]. One explanation for these differences may be differences in the definition of emergency presentation. Although several studies have used an algorithm to define emergency presentation [27], other studies regard emergency presentation as visiting an emergency service once before the diagnosis of CRC [27] or have used other criteria [25–26]. Our study defined emergency presentation as first patient contact with a hospital through an emergency service. Alternatively, GPs in Spain frequently refer patients with a high suspicion of cancer to emergency services to overcome limited access to investigations and outpatient appointments.

We found that the socio-demographic characteristics of these patients generally did not affect the rate of emergency presentation, although women with rectal, but not colon, cancer were more likely to use this route. Other studies have reported socio-demographic differences in emergency presentation, particularly in more vulnerable groups, including women, older patients, patients of low socioeconomic status and patients with higher comorbidity rates [17, 28–31]. These more vulnerable populations have less access to health services and usual care [32] and to colorectal examinations [33]. The discrepancy between our study and these other studies may be explained by the equal geographic distribution of primary care centers and hospitals in Spain. All groups would therefore have equal access to diagnostic examinations, specialists, and outpatient resources.

As shown in other studies, we observed that the type of initial symptom was highly associated with emergency presentation. Emergency presenters are more likely to present with abdominal pain [8,16,25,34], constipation [24,16,34], loss of weight [3, 34] fatigue and vomiting and significantly less likely to present with rectal bleeding [16,25,34,35] or changes in bowel habits [35]. Abdominal pain and abdominal obstruction are much more common in colon than in rectal cancer patients [8, 30], which may explain the higher rate of emergency presentation in those with colon cancer [3; 16, 17, 29]. Moreover, emergency presenters have symptoms of low predictive value, that is, not NICE-qualifying symptoms [25, 30, 36], and a higher number of symptoms. These findings suggest the need to review the role of symptoms with low predictive value for cancer when they appear simultaneously with other symptoms. However, emergency presenters do not report higher symptom severity, with similar proportions of emergency and non-emergency presenters contacting a doctor and disclosing symptoms to family and friends. Escalating or persisting symptoms, rather than severity of symptoms, could influence their visits to emergency services as described by others [37].

As previously reported [38], patients diagnosed with CRC by emergency services are quite heterogeneous. Some patients who experience symptoms go directly to an emergency service, whereas others visit their GP. The former group likely includes patients who postpone consultation for their symptoms, as well as those who respond immediately after experiencing disruptive symptoms, such as vomiting and abdominal pain [38].

Among patients who decided to visit a GP after symptom onset in the present work, fewer presented to emergency services, in agreement with other studies [25, 29–30]. Previous studies have found that, although similar proportions of emergency and non-emergency presenters consulted a primary-care physician [16, 25, 27], lower percentages of the former had ≥3 successive visits (25.2% vs. 39.2%) and referrals to outpatient consultations (36.3% vs. 76.0%). In addition, a higher proportion of emergency presenters were initially referred to an emergency service. These findings indicate that the diagnostic process in emergency presenters included fewer consultations and more referrals to emergency services by their GPs. Although it could be partially explained by the time from the clinical onset to diagnosis in these patients is inherently shorter and therefore the probability of having more consultations or referrals could be lower. Emergency presenters may have had more symptoms or more disruptive ones or may have been referred to emergency services to bypass long waiting lists or. Additionally some patients referred to an outpatient service may have contacted the hospital through an emergency service. Similarly, 30% of CRC patients in Scotland appropriately referred to secondary care were found to have been admitted to emergency services in the period between their referral and the appointment date [26].

Although emergency presentation has been reported to be a proxy of a delay in diagnosis [25], we found that the different intervals in the diagnostic pathway were shorter in emergency than in non-emergency presenters. These differences in symptom duration intervals may have been due to emergency presenters having more aggressive tumors, leading patients to quickly seek help and doctors to accelerate the testing schedules to arrive at a more rapid diagnosis. This may partly explain the waiting time paradox, in that patients with shorter times to diagnosis had higher mortality rates [39]. This does not mean that emergency admissions should be ignored but that the potential mortality benefits from a reduction may be less than hoped.

Finally, we found that patients registered in a teaching center were less likely to become emergency presenters than patients in non-teaching health services. Teaching health centers are characterized by better quality performance indicators and higher levels of educational activities, which may result in fewer referrals to emergency services. These findings are in agreement with a study showing that a higher total quality and outcomes framework protected against unplanned admissions [23].