Original article

Impact of faecal occult blood test screening on emergency admissions and short-term outcomes for colorectal cancer G. Libby1 , D. H. Brewster2 and R. J. C. Steele1 1

Medical Research Institute, University of Dundee, Dundee, and 2 Information Services Division, NHS National Services Scotland, Edinburgh, UK Correspondence to: Mrs G. Libby, Medical Research Institute, MBX7, Ninewells Hospital and Medical School, Dundee DD1 9SY, UK (e-mail: [email protected])

Background: Small studies have examined the effect of faecal occult blood test (FOBT) screening on the

proportion of hospital admissions for colorectal cancer (CRC) classed as an emergency. This study aimed to examine this and short-term outcomes in persons invited for screening compared with a control group not invited. Methods: The invited group comprised all individuals invited between 1 April 2000 and 31 July 2007 in the Scottish arm of the UK demonstration pilot of FOBT, and subsequently diagnosed with CRC aged 50–72 years between 1 May 2000 and 31 July 2009. The controls comprised all remaining individuals in Scotland not invited for FOBT but diagnosed with CRC aged 50–72 years in the same period. Results: There were 2981 people diagnosed with CRC in the group invited for screening (58⋅3 per cent participated) and 9842 in the control group. Multivariable regression adjusted for sex, age, deprivation, co-morbidities, tumour site and Dukes’ stage showed no difference between the groups for emergency admissions (odds ratio (OR) 0⋅89, 95 per cent confidence interval (c.i.) 0⋅77 to 1⋅02; P = 0⋅084) or length of hospital stay (LOS) (𝛃 coefficient −1⋅02 (95 per cent c.i. –1⋅05 to 1⋅01) days; P = 0⋅226). Comparing participants with controls, there were fewer emergency admissions (OR 0⋅59, 0⋅49 to 0⋅71; P < 0⋅001) and shorter LOS (𝛃 coefficient −1⋅06 (−1⋅10 to −1⋅02) days; P = 0⋅001). Short-term mortality was lower in the screened than the non-screened population (1⋅1 versus 2⋅8 per cent; P = 0⋅001). Conclusion: People who participated in FOBT screening had fewer emergency admissions and a shorter LOS. Deprivation was associated negatively with participation, but the impact of FOBT participation on emergency admissions was independent of deprivation level. The reduction in LOS has potential to reduce financial costs. Presented in part to United European Gastroenterology Week, Amsterdam, The Netherlands, October 2012 Paper accepted 11 June 2014 Published online 13 September 2014 in Wiley Online Library (www.bjs.co.uk). DOI: 10.1002/bjs.9613

Introduction

The major benefit of faecal occult blood test (FOBT) screening is the reduction in cause-specific mortality reported in randomized controlled trials1 and a population-based study2 , and shown to persist over 30 years of follow-up3 . The value of screening, however, can be enhanced if additional benefits are conferred; for example, a recent review4 highlighted how the introduction of a population screening programme can raise the quality of clinical practice and the present study examined the impact of FOBT screening on emergency hospital admissions for colorectal cancer (CRC). Emergency admissions are unpredictable, may disrupt hospital services, cause anxiety to patients and have © 2014 BJS Society Ltd Published by John Wiley & Sons Ltd

increased associated financial costs. Emergency admissions for CRC remain a substantial challenge5 – 7 . Such admissions occur more frequently in patients presenting with late-stage CRC8,9 . As FOBT screening aims to diagnose the disease at an early stage, it is anticipated that such screening will have an impact on emergency admissions. This has been examined previously in two small studies, which reported a reduction in emergency admissions in those participating in screening compared with a control group10 and an historical cohort not invited for screening11 . Participation in FOBT screening increases with decreasing deprivation and it could be argued that the reductions in emergency admissions in the above studies were explained by the screening test being completed by the less deprived, who are acknowledged to BJS 2014; 101: 1607–1615

1608

be more health conscious and to seek medical advice more readily, and therefore potentially be less likely to have an emergency admission. Deprivation may, therefore, be a confounding factor in the relationship between screening and emergency admissions, but this has not been examined previously. Co-morbidity may also confound this relationship if individuals with other health problems assign a low priority to screening and also delay seeking advice for symptoms, leading in turn to a later diagnosis of CRC12,13 . The present study therefore aimed to compare emergency admissions in a group invited for population-based FOBT screening with those in a group not invited, and accounting for deprivation and co-morbidity. In addition, length of hospital stay (LOS), as an indicator of financial cost (irrespective of type of admission), 30-day mortality and stoma formation were examined. Comparisons were also made between those from the invited group who participated in screening and the group not invited (control group). Methods

Ethical approval was not sought for the demonstration pilot of FOBT screening. This was a decision made by the National Screening Committee and endorsed by the UK Departments of Health on the grounds that screening for CRC using the FOBT is of proven efficacy, and the pilot constituted an evaluation of the feasibility of introducing a screening programme into the National Health Service (NHS). Permission to access and analyse the anonymized linked data presented in this paper was granted by the Community Health Index Advisory Board and the Privacy Advisory Committee, National Services Scotland, with additional permission from the Caldicott Guardian for the extra data in the All Scotland study.

Study population The study used data from two cohorts. The first, larger, cohort consisted of all people living in one of three Scottish health boards, who had been invited for FOBT screening in the Scottish arm of the UK demonstration pilot of FOBT14 and who had been subsequently diagnosed with CRC and, as a control group, people diagnosed with CRC in the remaining 11 area health boards who had not been invited during the pilot. The analysis of this cohort is referred to as the All Scotland study. A second, smaller, cohort was also used to examine the outcome of emergency admissions. This comprised individuals who had been diagnosed with CRC in a cohort originally compiled for the study examining the impact of FOBT on cause-specific mortality2 . This also comprised © 2014 BJS Society Ltd Published by John Wiley & Sons Ltd

G. Libby, D. H. Brewster and R. J. C. Steele

individuals invited as part of the Scottish arm of the UK pilot study, but they were individually matched for age, sex and deprivation with a control person drawn from a non-pilot health board. Deprivation was defined using the Scottish Index of Multiple Deprivation, an area-level measure that comprises seven indicators (employment, income, health, education, access to services, crime and housing) compiled from census and other data. Areas are ranked to produce a relative measure of deprivation, which is commonly expressed in quintiles of deprivation. The matched cohort had been compiled by NHS National Services Scotland in Edinburgh, using a population file of all individuals registered with a general practitioner in Scotland. However, there were 19 000 people who had been invited for screening, but could not be included in the cohort owing to lack of a unique matched control, and so the unmatched All Scotland cohort was used for the main analyses in the present study to avoid excluding this group. The group of people diagnosed with CRC from the matched study cohort was then used to carry out a sensitivity analysis for the outcome emergency admissions as deprivation is known to be associated strongly with participation in screening, and this group had been drawn from a cohort of invited and control pairs who had been matched originally for deprivation. The analysis of this cohort is henceforth referred to as the Matched Pilot study.

All Scotland study The invited-for-screening health board group comprised all people in Tayside, Grampian and Fife health boards invited for FOBT screening and diagnosed with CRC between 1 May 2000 (1 month after the first screening invitation) and 31 July 2009 (2 years after the last invitation in the pilot rounds) and aged 50–72 years at diagnosis. An alternative criterion would have been to include only people who were diagnosed within 2 years of a screening invitation, but this would have made it difficult to define a matching study interval for the control group. People diagnosed with CRC before the date of their first screening invitation were excluded. Following the successful completion of the pilot study, the Scottish FOBT screening programme was rolled out across all the Scottish health boards starting in 2007, and so the control health board group included everyone in the 11 non-pilot health boards diagnosed with CRC at age 50–72 years in the study interval (1 May 2000 to 31 July 2009), or to the date of the start of screening in their health board area if this was earlier. Cancer diagnoses were identified from the Scottish Cancer Registry and, using a unique patient identifier, were record-linked to hospital admission information (Scottish Morbidity Records, SMR01), date and cause of death www.bjs.co.uk

BJS 2014; 101: 1607–1615

Faecal occult blood test screening for colorectal cancer

(National Records of Scotland), and screening information (Bowel Screening database).

Definitions for study variables The admission of interest for each patient was either the admission during which the individual had been diagnosed with CRC or the first admission after their date of diagnosis (with no time limit). Patients were categorized by their type of admission as: initial treatment – the code for the ‘main operation’ variable in SMR01 for the admission of interest was included in a list of Classification of Interventions and Procedures, version 4 (OPCS-4) codes drawn up by the senior author (Table S1, supporting information); admission for CRC – the reason for the admission of interest was coded as CRC but the ‘main operation’ code was not in the list; or no admission – the date of diagnosis was not included in any admission and there were no admissions after the date of diagnosis that had CRC coded as the reason for admission. Hospital admissions are recorded on SMR01 by consultant episode, and even a short continuous inpatient stay can consist of a number of consultant episodes. For this study, an admission was defined as routine or emergency using the first consultant episode of the admission, even if the ‘initial treatment’ commenced later in the admission; the admission was classed as an emergency, for example, only if the initial consultant episode was recorded as an emergency. A consultant episode was defined as routine or emergency from SMR01 codes. The type of admission (routine or emergency) and LOS were identified only for those categorized as having had ‘initial treatment’. In addition, only people in the initial treatment subgroup were identified as having died within 30 days. The variable ‘any stoma’ was defined from a list of OPCS-4 codes drawn up by the senior author (Table S2, supporting information), and included both temporary and permanent stomas. An analysis was also carried out with exclusion of OPCS-4 H33.6 (anterior resection of rectum and exteriorization of bowel), a temporary procedure. An estimate of co-morbidity was made using two variables and adjusted for in the analyses. The first variable, previous admissions, was defined as the number of inpatient hospital admissions (for any reason) in the interval from 6 months to 5 years before the diagnosis of CRC. The second, duration of admission, was calculated as the total number of inpatient days for these admissions.

Statistical analysis The groups were assessed for differences in demographic measures using t tests and χ2 tests. The nature of screening © 2014 BJS Society Ltd Published by John Wiley & Sons Ltd

1609

is such that an increased number of cancers will be diagnosed, and so calculating emergency admissions as a percentage of cancers diagnosed may show a reduction in a group invited for screening owing to an inflated denominator even if there is no change in the number of emergency admissions. This was tested by calculating a rate ratio, standardized by age and sex, of the observed number of emergency admissions in the invited group compared with the number expected if the rates of emergency admission in the control group were applicable. The rates used the total estimated mid-year populations for the interval 2000–2009 for the three pilot health boards for the invited group and the remaining health boards for the control group. Differences in the relationship with the outcome measures between the invited-for-screening and control health board groups were examined using regression methods. Factors considered relevant to screening were each assessed for an independent association with the outcomes, but the final model was not driven by the statistical significance of these associations as all the variables were either already known to be associated with screening or being tested in the analyses. Logistic regression was used for the outcomes emergency admissions and stoma formation. Linear regression was used to examine LOS after the distribution of the variable had been assessed by histogram and the Shapiro–Wilk test of normality. Potential interactions between variables in the regression analyses were tested using the likelihood ratio test. P < 0⋅050 was considered statistically significant, but P values were also interpreted in the context of the number of subjects being tested. All analyses were carried out using Stata® version 10.0 (StataCorp LP, College Station, Texas, USA). Results

All Scotland study A total of 3166 people were diagnosed with CRC in the invited-for-screening health board group during the study period; 185 of these were diagnosed before their first date of invitation, leaving 2981 for inclusion in the study. Some 9842 people were diagnosed in the same interval in the control group. The demographic characteristics of the two groups are shown in Table 1 together with those of the participants in the invited-for-screening health board group. Comparisons of the invited and control health board groups showed no difference in mean age, although the control group had a slightly higher proportion in both the youngest and oldest subgroups. There was a difference in the distribution of deprivation quintiles between the www.bjs.co.uk

BJS 2014; 101: 1607–1615

1610

G. Libby, D. H. Brewster and R. J. C. Steele

groups; this difference remained even if Greater Glasgow and Clyde health board, in which 38⋅9 per cent of people were in the highest deprivation quintile, was excluded from the study population. In both groups, there was an increasing trend in the number of previous admissions and duration of previous admissions with increasing deprivation. A higher proportion of patients in the invited health board group had ‘initial treatment’ for CRC (72⋅9 versus 65⋅4 per cent). There was a smaller percentage of emergency admissions in the invited group compared with controls (Table 1). Calculating the standardized rate ratio using observed Table 1

emergency admissions of 357 in the invited group and expected admissions of 432⋅8 produced a value of 82⋅5 (95 per cent confidence interval (c.i.) 74⋅3 to 91⋅5). As this also indicated a reduction in the invited group, there appeared to be no evidence to suggest that using the percentage of emergency admissions and logistic regression was not appropriate. Of the 2981 people who had been invited for screening in the invited health board group, 1739 (58⋅3 per cent) had participated in the most recent screening round before diagnosis and 64⋅6 per cent (an additional 186 people) had participated in at least one round before diagnosis.

Demographics of cohort for All Scotland study

Age at diagnosis (years)* Age group at diagnosis 50–54 55–59 60–64 65–69 ≥ 70 Men SIMD 1 (high deprivation) 2 3 4 5 (low deprivation) Not known No. of previous admissions* Duration of admission (days)* Participated Yes No Tumour site Colon Rectum Dukes’ stage A B C ‘D’ Not known Type of admission Initial treatment CRC admission No admission Urgency of admission† Routine admission Emergency admission Stoma formation† Any stoma Temporary stoma‡ excluded Length of hospital stay (days)*† Died ≤ 30 days after initial treatment†

Participants (n = 1739)

Invited health boards (n = 2981)

Control health boards (n = 9842)

64⋅0(5⋅6)

63⋅9(5⋅6)

63⋅8 (0⋅59)

116 (6⋅7) 292 (16⋅8) 430 (24⋅7) 574 (33⋅0) 327 (18⋅8) 986 (56⋅7)

215 (7⋅2) 498 (16⋅7) 764 (25⋅6) 942 (31⋅6) 562 (18⋅9) 1764 (59⋅2)

873 (8⋅9) 1584 (16⋅1) 2273 (23⋅1) 3114 (31⋅6) 1998 (20⋅3) 5827 (59⋅2)

168 (9⋅7) 273 (15⋅7) 385 (22⋅1) 451 (25⋅9) 462 (26⋅6) 0 (0) 0⋅75(1⋅73) 3⋅8(10⋅8)

367 (12⋅3) 509 (17⋅1) 656 (22⋅0) 734 (24⋅6) 715 (24⋅0) 0 (0) 0⋅77(1⋅75) 4⋅2(13⋅8)

2225 (22⋅6) 2258 (22⋅9) 2016 (20⋅5) 1610 (16⋅4) 1732 (17⋅6) 1 (0.0) 0⋅85(1⋅87) 4⋅9(15⋅4)

– –

1739 (58⋅3) 1242 (41⋅7)

– –

1095 (63.0) 644 (37⋅0)

1844 (61⋅9) 1137 (38⋅1)

6111 (62⋅1) 3731 (37⋅9)

379 (21⋅8) 421 (24⋅2) 478 (27⋅5) 210 (12⋅1) 251 (14⋅4)

497 (16⋅7) 774 (26⋅0) 813 (27⋅3) 459 (15⋅4) 438 (14⋅7)

1169 (11⋅9) 2600 (26⋅4) 2513 (25⋅5) 1998 (20⋅3) 1562 (15⋅9)

1316 (75⋅7) 292 (16⋅8) 131 (7⋅5)

2172 (72⋅9) 575 (19⋅3) 234 (7⋅8)

6439 (65⋅4) 2488 (25⋅3) 915 (9⋅3)

1166 (88⋅6) 150 (11⋅4)

1815 (83⋅6) 357 (16⋅4)

5180 (80⋅4) 1259 (19⋅6)

225 (17⋅1) 157 (11⋅9) 13⋅8(9⋅4) 15 (1⋅1)

443 (20⋅4) 341 (15⋅7) 14⋅8(11⋅4) 31 (1⋅4)

1514 (23⋅5) 1219 (18⋅9) 15⋅7(14⋅2) 182 (2⋅8)

P§ 0⋅647¶ 0⋅002

0⋅976 < 0⋅001

0⋅050¶ 0⋅039¶

0⋅819

< 0⋅001

< 0⋅001

0⋅001

0⋅003 0⋅001 0⋅004¶ < 0⋅001

Values in parentheses are percentages unless indicated otherwise; *values are mean(s.d.). †Data for patients in initial treatment category. ‡Anterior resection of rectum and exteriorization of bowel. SIMD, Scottish Index of Multiple Deprivation; CRC, colorectal cancer. §χ2 test, except ¶t test (invited health boards versus control health boards).

© 2014 BJS Society Ltd Published by John Wiley & Sons Ltd

www.bjs.co.uk

BJS 2014; 101: 1607–1615

Faecal occult blood test screening for colorectal cancer

1611

The former group was used to compare participants and the control health board group for emergency admissions, stoma formation and LOS.

Stoma formation Unadjusted logistic regression analyses showed that male sex, increasing deprivation, rectal cancer and more advanced disease were all associated with increasing stoma formation (including permanent and temporary stomas) (Table 2). However, stoma formation was less common in the group invited for screening, with little difference between the unadjusted and adjusted analyses (adjusted OR 0⋅85, 95 per cent c.i. 0⋅74 to 0⋅97; P = 0⋅016). The adjusted OR for the analysis with exclusion of temporary stoma (H33⋅6) was very similar (OR 0⋅84, 0⋅73 to 0⋅97; P = 0⋅017).

Emergency admissions Unadjusted logistic regression analyses showed that emergency admissions were associated with female sex, increasing deprivation and later stage at diagnosis (Table 2). Tumour site was also associated with emergency admissions, which were significantly less frequent in those diagnosed with rectal cancer. There were fewer emergency admissions in the invited health boards compared with the control health boards, and this remained statistically significant when adjusted for age, sex, previous admissions, duration of previous admissions, tumour site and Dukes’ stage (odds ratio (OR) 0⋅86, 95 per cent c.i. 0⋅75 to 0⋅98; P = 0⋅027), but not when additionally adjusted for deprivation (OR 0⋅89, 0⋅77 to 1⋅02; P = 0⋅084). Likelihood ratio tests showed a significant interaction between the two groups and Dukes’ stage, but not between group and deprivation or Dukes’ stage and deprivation. To examine the interaction between group and Dukes’ stage, the adjusted analysis was repeated for each Dukes’ stage separately; this showed significantly fewer emergency admissions in the invited health board group for people diagnosed at Dukes’ stage A only (OR 0⋅13, 0⋅04 to 0⋅43; P = 0⋅001). Three people in the invited health board group and 48 in the control group were diagnosed with Dukes’ stage A disease. Table 2

Length of hospital stay A histogram of LOS showed a skewed distribution, which was confirmed by the Shapiro–Wilk test for normality. Leaving the data untransformed was considered owing to the size of the cohort, but it was decided to log transform the data and a subsequent histogram showed a distribution close to normal. Linear regression was therefore used and the data transformed back after analysis, giving the results in the original units of ‘days’. Unadjusted linear regression showed that LOS was associated with increasing age, increasing deprivation, previous admissions, rectal cancer and later stage at diagnosis (Table 2). LOS was shorter by 1 day in the group invited for screening, but did not remain statistically significant when adjusted for all other variables: β coefficient −1⋅02 (95 per cent c.i. – 1⋅05 to 1⋅01) days (P = 0⋅226).

All Scotland study: logistic regression with emergency admission, any stoma and length of stay as outcome Emergency admission Odds ratio

Unadjusted analysis Sex (F versus M) Age at diagnosis SIMD 2 versus 1 (high deprivation) 3 versus 1 4 versus 1 5 versus 1 Previous admissions Duration of previous admission Tumour site (rectum versus colon) Dukes’ stage B versus A C versus A ‘D’ versus A Not known versus A Invited versus control health board Participants versus control health board Adjusted analysis* Invited versus control health board Participants versus control health board

Any stoma P

Length of hospital stay

Odds ratio

P

β coefficient (days)

P

1⋅12 (1⋅01, 1⋅25) 1⋅00 (0⋅97, 1⋅06)

0⋅037 0⋅520

0⋅63 (0⋅57, 0⋅70) 0⋅98 (0⋅98, 0⋅99)

< 0⋅001 0⋅001

−1⋅00 (−1⋅03, 1⋅02) 1⋅01 (1⋅01, 1⋅01)

0⋅825 < 0⋅001

0⋅87 (0⋅74, 1⋅03) 0⋅79 (0⋅66, 0⋅93) 0⋅73 (0⋅62, 0⋅87) 0⋅77 (0⋅64, 0⋅91) 1⋅00 (0⋅97, 1⋅04) 1⋅00 (0⋅99, 1⋅01) 0⋅32 (0⋅28, 0⋅36)

0⋅098 0⋅005 0⋅001 0⋅002 0⋅817 0⋅450 < 0⋅001

0⋅92 (0⋅79, 1⋅07) 0⋅86 (0⋅74, 1⋅01) 0⋅84 (0⋅71, 0⋅99) 0⋅78 (0⋅66, 0⋅92) 0⋅97 (0⋅94, 1⋅00) 0⋅99 (0⋅99, 1⋅00) 8⋅33 (7⋅41, 9⋅38)

0⋅293 0⋅065 0⋅034 0⋅003 0⋅052 0⋅142 < 0⋅001

−1⋅03 (−1⋅07, 1⋅01) −1⋅07 (−1⋅11, −1⋅03) −1⋅11 (−1⋅15, −1⋅06) −1⋅13 (−1⋅17, −1⋅08) 1⋅02 (1⋅01, 1⋅03) 1⋅00 (1⋅06, 1⋅12) 1⋅09 (1⋅06, 1⋅12)

0⋅132 0⋅001 < 0⋅001 < 0⋅001 < 0⋅001 < 0⋅001 < 0⋅001

4⋅54 (3⋅37, 6⋅12) 6⋅39 (4⋅76, 8⋅59) 12⋅33 (9⋅07, 16⋅76) 5⋅08 (3⋅64, 7⋅10) 0⋅81 (0⋅77, 0⋅92) 0⋅53 (0⋅44, 0⋅63)

< 0⋅001 < 0⋅001 < 0⋅001 < 0⋅001 0⋅001 < 0⋅001

0⋅75 (0⋅64, 0⋅88) 0⋅83 (0⋅71, 0⋅98) 1⋅22 (1⋅01, 1⋅47) 1⋅85 (1⋅52, 2⋅25) 0⋅83 (0⋅74, 0⋅94) 0⋅67 (0⋅57, 0⋅78)

0⋅001 0⋅024 0⋅035 < 0⋅001 0⋅003 < 0⋅001

1⋅09 (1⋅06, 1⋅14) 1⋅11 (1⋅07, 1⋅16) 1⋅09 (1⋅05, 1⋅15) 1⋅13 (1⋅07, 1⋅19) −1⋅04 (−1⋅07, −1⋅01) −1⋅09 (−1⋅13, −1⋅05)

< 0⋅001 < 0⋅001 < 0⋅001 < 0⋅001 0⋅009 < 0⋅001

0⋅89 (0⋅77, 1⋅02) 0⋅59 (0⋅49, 0⋅71)

0⋅084 < 0⋅001

0⋅85 (0⋅74, 0⋅97) 0⋅67 (0⋅57, 0⋅79)

0⋅016 < 0⋅001

−1⋅02 (−1⋅05, 1⋅01) −1⋅06 (−1⋅10, −1⋅02)

0⋅226 0⋅001

Values in parentheses are 95 per cent confidence intervals. SIMD, Scottish Index of Multiple Deprivation. *Adjusted for all variables.

© 2014 BJS Society Ltd Published by John Wiley & Sons Ltd

www.bjs.co.uk

BJS 2014; 101: 1607–1615

1612

G. Libby, D. H. Brewster and R. J. C. Steele

Mortality at 30 days A χ2 test was used to compare 30-day mortality in the invited and control health board groups; this showed lower mortality in the invited group (1⋅4 versus 2⋅8 per cent; P < 0⋅001).

Comparison of screening participants and controls Logistic regression showed that participants were less likely to have an emergency admission or stoma formation than those in the control health board group, in both the unadjusted and adjusted analyses (emergency admissions: adjusted OR 0⋅59, 95 per cent c.i. 0⋅49 to 0⋅71, P < 0⋅001; any stoma: adjusted OR 0⋅67, 0⋅57 to 0⋅79, P < 0⋅001) (Table 2). Linear regression analysis showed a shorter LOS Table 3

in participants, both unadjusted and adjusted for all other variables: adjusted β coefficient −1⋅06 (95 per cent c.i. –1⋅10 to −1⋅02) days (P = 0⋅001). Fewer participants had died within 30 days of initial treatment compared with people in the control health board group (1⋅1 versus 2⋅8 per cent; P = 0⋅001).

Matched pilot study The variable definitions and methods were the same in this study as for the All Scotland study, although no one in the invited group had been diagnosed before their first invitation for screening. The demographic characteristics of the invited and control health board groups are shown in Table 3, together with those in the invited group who

Demographics of matched pilot study cohort

Age at diagnosis (years)* Age group at diagnosis 50–54 55–59 60–64 65–69 ≥ 70 Men SIMD 1 (high deprivation) 2 3 4 5 (low deprivation) No. of previous admissions* Duration of admission (days)* Tumour site Colon Rectum Dukes’ stage A B C ‘D’ Not known Type of admission Initial surgery CRC admission No admission Urgency of admission† Routine admission Emergency admission Stoma formation† Any stoma Temporary stoma‡ excluded Length of hospital stay (days)*† Died ≤ 30 days after initial treatment†

Participants (n = 1716)

Invited health boards (n = 2626)

Control health boards (n = 2356)

63⋅7(5⋅5)

63⋅6(5⋅6)

63⋅8(5⋅67)

147 (8⋅6) 298 (17⋅4) 464 (27⋅0) 594 (34⋅6) 213 (12⋅4) 1012 (52⋅8)

242 (9⋅2) 468 (17⋅8) 691 (26⋅3) 874 (33⋅3) 351 (13⋅4) 1612 (61⋅4)

168 (7⋅1) 408 (17⋅3) 591 (25⋅1) 730 (31⋅0) 459 (19⋅5) 1445 (61⋅3)

178 (10⋅4) 276 (16⋅1) 389 (22⋅7) 442 (25⋅8) 431 (25⋅1) 0⋅72(1⋅71) 3⋅6(10⋅3)

353 (13⋅4) 442 (16⋅8) 580 (22⋅1) 644 (24⋅5) 607 (23⋅1) 0⋅74(1⋅72) 3⋅9(11⋅8)

324 (13⋅8) 425 (18⋅0) 520 (22⋅1) 573 (24⋅3) 514 (21⋅8) 0⋅83(1⋅88) 4⋅8(15⋅0)

1081 (63⋅0) 635 (37⋅0)

1625 (61⋅9) 1001 (38⋅1)

1448 (61⋅5) 908 (38⋅5)

374 (21⋅8) 412 (24⋅0) 457 (26⋅6) 231 (13⋅5) 242 (14⋅1)

450 (17⋅1) 680 (25⋅9) 702 (26⋅7) 415 (15⋅8) 379 (14⋅4)

273 (11⋅6) 633 (26⋅9) 622 (26⋅4) 457 (19⋅4) 371 (15⋅7)

1301 (75⋅8) 287 (16⋅7) 128 (7⋅5)

1939 (73⋅8) 490 (18⋅7) 197 (7⋅5)

1614 (68⋅5) 542 (23⋅0) 200 (8⋅5)

1136 (87⋅3) 165 (12⋅7)

1607 (82⋅9) 331 (17⋅1)

1307 (81⋅0) 307 (19⋅0)

229 (17⋅6) 165 (12⋅7) 14⋅1(9⋅6) 13 (1⋅0)

398 (20⋅5) 306 (15⋅8) 15⋅4(21⋅0) 30 (1⋅5)

372 (23⋅0) 307 (19⋅0) 15⋅0(11⋅5) 38 (2⋅4)

P§ 0⋅171¶ < 0⋅001

0⋅969 0⋅721

0⋅087¶ 0⋅019¶ 0⋅760

< 0⋅001

< 0⋅001

0⋅133

0⋅069 0⋅011 0⋅099¶ 0⋅086

Values in parentheses are percentages unless indicated otherwise; *values are mean(s.d.). †Data for patients in initial treatment category. ‡Anterior resection of rectum and exteriorization of bowel. SIMD, Scottish Index of Multiple Deprivation; CRC, colorectal cancer. §χ2 test, except ¶t test (invited health boards versus control health boards).

© 2014 BJS Society Ltd Published by John Wiley & Sons Ltd

www.bjs.co.uk

BJS 2014; 101: 1607–1615

Faecal occult blood test screening for colorectal cancer

Table 4

1613

Matched pilot study: logistic regression with emergency admission as outcome Unadjusted analysis Odds ratio

Sex (F versus M) Age at diagnosis SIMD 2 versus 1 (high deprivation) 3 versus 1 4 versus 1 5 versus 1 Previous admissions Duration of previous admission Tumour site (rectum versus colon) Dukes’ stage B versus A C versus A ‘D’ versus A Not known versus A Invited versus control health board Participants versus control health board

Adjusted analysis* P

1⋅23 (1⋅04, 1⋅47) 0⋅99 (0⋅98, 1⋅01)

0⋅017 0⋅719

0⋅81 (0⋅60, 1⋅09) 0⋅78 (0⋅59, 1⋅04) 0⋅82 (0⋅61, 1⋅08) 0⋅68 (0⋅51, 0⋅90) 0⋅97 (0⋅91, 1⋅02) 0⋅99 (0⋅98, 1⋅01) 0⋅31 (0⋅25, 0⋅39)

0⋅174 0⋅094 0⋅152 0⋅008 0⋅231 0⋅572 < 0⋅001

8⋅09 (4⋅66, 14⋅07) 10⋅30 (5⋅95, 17⋅84) 20⋅63 (11⋅70, 36⋅37) 7⋅99 (4⋅32, 14⋅96) 0⋅88 (0⋅74, 1⋅04) 0⋅62 (0⋅50, 0⋅76)

< 0⋅001 < 0⋅001 < 0⋅001 < 0⋅001 0⋅134 < 0⋅001

Odds ratio

P

0⋅94 (0⋅78, 1⋅12) 0⋅66 (0⋅53, 0⋅82)

0⋅489 < 0⋅001

Values in parentheses are 95 per cent confidence intervals. SIMD, Scottish Index of Multiple Deprivation. *Adjusted for all variables.

participated. There were some differences compared with the All Scotland study; most notably, the distribution of deprivation quintiles was similar in the two groups in the matched pilot study. In addition, there was a higher proportion of men in this smaller cohort, probably owing to the excluded 19 000 people comprising a majority of women. There was no difference in 30-day mortality between the groups.

for sex, tumour site and Dukes’ stage, the only variables showing a significant association with emergency admissions in the unadjusted analysis of the participants and control health board cohort. Dukes’ stage and rectal cancer remained independently associated with emergency admissions in the adjusted analysis.

Emergency admissions: comparison of invited and control groups This cohort was used to carry out a sensitivity analysis for the outcome emergency admissions as the individuals diagnosed with CRC had been drawn from a cohort of people invited for screening and controls who had been matched originally by deprivation quintile. The results from the logistic regression analyses are shown in Table 4. In this cohort, female sex and more advanced disease were positively associated with emergency admissions, whereas patients with rectal cancer had fewer emergency admissions. Deprivation level was not generally associated with emergency admissions, and there was no difference between the invited and control health board groups.

In this study there was no statistically significant difference in either admissions for CRC classified as an emergency or LOS between individuals invited for FOBT screening and those not invited. In contrast, there were fewer emergency admissions and a shorter LOS in those who participated in screening compared with the controls, a group of people who, if invited for screening, would have comprised both participants and non-participants. Stoma formation was significantly less common in those invited compared with controls, and was reduced even further in those who participated possibly owing to the impact of screening on earlier stage at diagnosis when stoma formation is less common. A reduction in 30-day mortality was also reported, both in those invited and in those who participated compared with the control group. These results could not be adjusted for potential confounders because of the small number of deaths. The different results for the proportion of emergency admissions in participants and the invited-for-screening group as a whole could suggest an effect of selection bias. Health consciousness and health-seeking behaviour, including participation in screening, are known to be more common in the less deprived and it is possible that this

Emergency admissions: comparison of screening participants and controls Comparing participants in screening with the control health board group showed significantly fewer emergency admissions among participants, with similar ORs in both unadjusted and adjusted analyses (adjusted OR 0⋅66, 95 per cent c.i. 0⋅53 to 0⋅82; P < 0⋅001). The adjusted OR remained the same whether adjusted for all variables, or © 2014 BJS Society Ltd Published by John Wiley & Sons Ltd

Discussion

www.bjs.co.uk

BJS 2014; 101: 1607–1615

1614

G. Libby, D. H. Brewster and R. J. C. Steele

factor influenced the risk of an emergency admission rather than the process of screening itself. Data from both study cohorts do not, however, support this idea. Although there was an increase in participation with decreasing deprivation, participation was not limited to the less deprived as half of the two most deprived groups invited for screening in the All Scotland study, and more than half in the matched pilot study, participated. In addition, although the adjusted regression analysis revealed an independent association between deprivation and emergency admissions, it also found an independent association between participation and a reduced likelihood of emergency admissions. These results therefore suggest that the benefit of screening derives from completing the screening test rather than background level of deprivation, with very limited evidence of selection bias. The benefits of population-based screening for CRC, including a reduction in emergency hospital admissions, obviously depend on the level of participation. In the All Scotland study, which used the guaiac FOBT, the participation rate was 58⋅3 per cent overall and increasing this is a priority. Introducing other screening modalities suitable for population-based screening may lead to improved participation15 . The impact of FOBT screening on emergency admissions, stoma formation and 30-day mortality has been examined previously in two studies using different study designs. The first used data from a randomized controlled trial of FOBT10 that was designed to assess the impact of FOBT screening on CRC mortality; a direct comparison was made between emergency admissions, 30-day mortality rate and stoma formation in those randomized to be invited for screening and those not invited. The results from this trial also reported a reduction in emergency admissions only among those who were invited and subsequently participated in screening compared with the control group (5⋅3 versus 27⋅9 per cent). Reductions in stoma formation and 30-day mortality were also reported, but again only for participants. The second study11 used data from University Hospitals Coventry and Warwickshire NHS Trust within the English arm of the UK bowel cancer screening pilot. It examined the proportion of hospital admissions for CRC classified as an emergency from the year before FOBT screening was introduced through 5 subsequent years of screening. The results also indicated a reduction in emergency CRC admissions for the screening population, from 29⋅3 per cent before screening to 15⋅8 per cent 5 years later. However, further data also appeared to show a reduction in emergency admissions in the same interval for patients diagnosed above the screening age, which may indicate that factors other

than screening influenced emergency admissions in this population16 . Therefore, it is not possible to compare directly the results from this study with those from the All Scotland study. The role of co-morbidities on the impact of FOBT screening on emergency admissions was also considered in the present study. However, the measures of co-morbidity appeared to have no independent effect on the level of emergency admissions. They were shown to increase with increasing deprivation in both the invited-for-screening and control groups, and this may explain the lack of any additional, independent effect. A more robust measure, such as the Charlson Co-morbidity Index, may have shown an effect as patients presenting as an emergency are often in a poorer physical condition17 . However, the index was not compiled for this study. Admissions to hospital for emergency CRC surgery are associated with a longer LOS than elective admissions17,18 , and it would seem reasonable to assume that the impact of FOBT screening on emergency admissions will also affect LOS. In this study, the impact of FOBT screening on LOS was examined, irrespective of type of admission, as an indicator of financial cost. There was no difference between the invited and control groups, but there was a reduction of 1 day in the LOS for those who participated compared with controls. This appears a modest reduction but inpatient hospital costs associated with CRC can be substantial, and at the population level even this reduction in duration of inpatient stay can produce financial savings19 . The notable strength of this study was the size of the study cohort, with over 12 000 patients diagnosed with CRC, compared with 1962 and 1215 patients in the two previous studies10,11 . The study was also able to assess the outcomes in a true population-based programme of FOBT screening, but with a more robust study design than previous investigations. It was able to take account of deprivation and co-morbidities in the relationship between FOBT screening and emergency admissions, and additionally assess the impact on LOS for the first time in population-level screening. Although an observational study, with recognized limitations on the level of matching between groups being compared, this remains an appropriate study design to assess outcomes in the real world.

© 2014 BJS Society Ltd Published by John Wiley & Sons Ltd

www.bjs.co.uk

Acknowledgements

This work was supported by a grant from the Chief Scientist Office (grant no. CZH/6/4), Scottish Government Health Directorates, to establish a bowel screening research unit. Disclosure: The authors declare no conflict of interest.

BJS 2014; 101: 1607–1615

Faecal occult blood test screening for colorectal cancer

1615

References 1 Hewitson P, Glasziou P, Watson E, Towler B, Irwig L. Cochrane systematic review of colorectal cancer screening using the fecal occult blood test (Hemoccult): an update. Am J Gastroenterol 2008; 103: 1541–1549. 2 Libby G, Brewster DH, McClements PL, Carey FA, Black RJ, Birrell J et al. The impact of population-based faecal occult blood test screening on colorectal cancer mortality: a matched cohort study. Br J Cancer 2012; 107: 255–259. 3 Shaukat A, Mongin SJ, Geisser MS, Lederle FA, Bond JH, Mandel JS et al. Long-term mortality after screening for colorectal cancer. N Engl J Med 2013; 369: 1106–1114. 4 Bretthauer M, Kalager M. Principles, effectiveness and caveats in screening for cancer. Br J Surg 2013; 100: 55–65. 5 Health and Social Care Information Centre in partnership with the Association of Coloproctology of Great Britain and Ireland and the Royal College of Surgeons. National Bowel Cancer Audit Annual Report 2013. https://catalogue.ic.nhs.uk/ publications/clinical/bowel/nati-clin-audi-supp-prog-bowecanc-2013/nati-clin-audi-supp-prog-bowe-canc-2013-rep1. pdf [accessed 1 December 2013]. 6 Barnett A, Cedar A, Siddiqui F, Herzig D, Fowlkes E, Thomas CR Jr. Colorectal cancer emergencies. J Gastronintest Canc 2013; 44: 132–142. 7 Esteva M, Leiva A, Ramos M, Pita-Fernández S, González-Luján L, Casamitjana M et al.; DECCIRE GROUP. Factors related with symptom duration until diagnosis and treatment of symptomatic colorectal cancer. BMC Cancer 2013; 13: 87. 8 Mella J, Biffen A, Radclifffe AG, Stamatakis JD, Steele RJ. A population based audit of colorectal cancer management in two United Kingdom health districts. Colorectal Cancer Working Group, Royal College of Surgeons of England Clinical Epidemiology and Audit Unit. Br J Surg 1997; 84: 1731–1736. 9 McPhail S, Elliss-Brookes L, Shelton J, Ives A, Greensson S,

10

11

12

13

14

15

16 17

18

19

Vernon S. Emergency presentation of cancer and short term mortality. Br J Cancer 2013; 109: 2027–2034. Scholefield JH, Robinson MH, Mangham CM, Hardcastle JD. Screening for colorectal cancer reduces emergency admissions. Eur J Surg Oncol 1998; 24: 47–50. Goodyear SJ, Leung E, Menon A, Pedamallu S, Williams N, Wong LS. The effects of population-based faecal occult blood test screening upon emergency colorectal cancer admissions in Coventry and North Warwickshire. Gut 2008; 57: 218–222. Miles A, Rainbow S, Von Wagner C. Cancer fatalism and poor self-related health mediate the association between socioeconomic status and uptake of colorectal cancer screening in England. Cancer Epidemiol Biomarkers Prev 2011; 20: 2132–2140. Sutton S, Wardle J, Taylor T, McCaffery K, Williamson D, Edwards R et al. Predictors of attendance in the United Kingdom flexible sigmoidoscopy screening trial. J Med Screen 2000; 7: 99–104. Steele RJ, McClements PL, Libby G, Black R, Morton C, Birrell J et al. Results from the first three rounds of the Scottish demonstration pilot of FOBT screening for colorectal cancer. Gut 2009; 58: 530–535. Digby J, McDonald PJ, Strachan JA, Libby G, Steele RJ, Fraser CG. Use of the faecal immunochemical test narrows current gaps in uptake for sex, age and deprivation in a bowel cancer screening programme. J Med Screen 2013; 20: 80–85. Goodyear SJ. Letter: Author’s response. Gut 2008; 57: 1333–1334. Cuffy M, Abir F, Audisio RA, Longo WA. Colorectal cancer presenting as surgical emergencies. Surg Oncol 2004; 13: 149–157. McArdle CS, Hole DJ. Emergency presentation of colorectal cancer associated with poor 5-year survival. Br J Surg 2004; 91: 605–609. Kelly M, Sharp L, Dwane F, Kelleher T, Comber H. Factors predicting hospital length-of-stay and readmission after colorectal resection: a population-based study of elective and emergency admissions. BMC Health Serv Res 2012; 12: 77.

Supporting information

Additional supporting information may be found in the online version of this article: Table S1 OPCS-4 codes for ‘initial treatment’ (Word document) Table S2 OPCS-4 codes for stoma (Word document)

© 2014 BJS Society Ltd Published by John Wiley & Sons Ltd

www.bjs.co.uk

BJS 2014; 101: 1607–1615