Colorectal cancer (CRC) constitutes the second highest cause of cancer mortality and third most commonly diagnosed cancer in Canada [1]. In 2013, the incidence of CRC in Canada is projected to be 60 per 100,000 in males and 40 per 100,000 in females, with approximately 4200 women and 5000 men dying from it [1]. The demographic shift towards an older population has increased CRC incidence and mortality over the last thirty years [2]. Despite opportunities for early detection [2], CRC tends to be diagnosed at advanced cancer stages [3], which are associated with both lower probabilities of survival [4] and greater costs to the healthcare system [5].

CRC screening guidelines have been developed and adapted based on the recognition that early detection and treatment can significantly reduce CRC morbidity and mortality [6]. These guidelines were based on existing knowledge of the best means for early detection of cancer [7–9] (since updated [10]) with screening recommendations typically calling for annual or biennial screening by a noninvasive stool-based test such as a fecal occult blood test (FOBT) with follow-up with endoscopic tests such as flexible sigmoidoscopy or colonoscopy. A Cochrane systematic review of several large randomized trials approximating population-based screening strategies incorporating biennial fecal occult blood tests estimated that screening reduces CRC mortality by 15% [11]. Population-based screening is generally thought to be cost effective [12, 13] and is supported by Canadian [14, 15] and other [16] microsimulation models.

Screening rates for CRC in Canada remain quite low despite the presence of screening guidelines, evidence of screening effectiveness, and universal health insurance coverage for these procedures [17]. In 2003, adherence to screening guidelines for FOBTs among average risk, target-age individuals was only 15% [18], increasing to 23% in 2008 [19]. Screening rates have also been observed to vary by province, with Ontario historically having higher rates than much of the rest of Canada [19]. By contrast, 73% of Canadian women aged 50–69 reported mammogram adherence in this time period [20]. Evidence suggests that both Canadian patients [21] and physicians [21, 22] were unknowledgeable of, or non-adherent to, recommended screening protocols for CRC. Outside of Ontario and in Ontario prior to 2008, CRC screening broadly followed the procedures as outlined in the national guidelines [7–9], albeit in an opportunistic or individualistic manner which gave rise to concerns of underutilized or inappropriate screening and lack of quality assurance [23].

In order to address low adherence to screening guidelines, Ontario’s Ministry of Health and Long Term Care launched an organized population-based colorectal cancer screening program called ColonCancerCheck in March of 2008 [23]. ColonCancerCheck targeted adults ages 50 and older with the goals of reducing CRC mortality and enhancing the involvement of physicians in patient screening. The program utilized existing screening recommendations for biennial FOB testing with endoscopy follow-up for average risk adults with colonoscopy being recommended for those of high risk. It also introduced outreach and organizational frameworks designed to increase screening adherence and educate both the public and health care providers about proper screening protocols.

To increase adherence to screening practices in the target population, ColonCancerCheck directed educational media campaigns to physicians through information kits and counseling manuals, and to the public, through television, websites, posters, pamphlets, and street teams at public events. With primary care providers (PCPs) taking the central role of distributing kits and following up with patients, arrangements were made for those without PCPs to obtain kits through pharmacies or through calling a 1–800 number. Contracts were established with laboratories to process tests and develop quality-control standards. A single brand of FOBT kit with sufficient accuracy and modest dietary restrictions was selected and used by all participating laboratories to help ensure quality, consistency, and interpretability of results. FOBT kits were handed out with instructions and post-paid envelopes to be completed at home and mailed to central laboratories for processing. An organizational framework was developed for screening invitations and results letters with pilot programs for invitation-based outreach in development. PCPs would be informed of their patient’s test results and follow-through by arranging colonoscopies if positive, with negative results leading to notification and recall for testing in two years. Incomplete and unreadable tests would be corrected with instructions returned for re-tests [23]. Prior to the launch of the program, FOBT kits were sent out to over 10,000 physician and nurse practitioner offices and 3000 pharmacies [23].

Some evidence suggests increased screening in Ontario after the program’s introduction. Health insurance claims indicate that 30% of Ontarians aged 50–74 had at least one complete FOBT in the previous two years in 2007–8 (inclusive) compared to 20% in 2005–6 [23]. Survey data corroborates this, suggesting FOBT adherence rose from 21% to 31% from 2005 to 2008, while endoscopy test adherence rose from 24% in 2005 to 30% in 2008, in those 50 and older [19]. However, these findings are insufficient to draw conclusions concerning the impact of ColonCancerCheck since there was a general trend of increased screening across Canada with increases observed in several provinces without screening programs during this time period [19]. Similarly designed programs were under various stages of development in Manitoba [24] and Alberta [25] during this time, with Manitoba instigating a pilot program for CRC screening in two health regions. However, no province-wide CRC screening programs outside of Ontario were implemented during the study period.

The purpose of the present study was to measure and describe the impact of ColonCancerCheck on CRC screening in the asymptomatic, average risk population in Ontario. We used a quasi-experimental design and a difference-in-differences model, which measures the change in screening rates in Ontario relative to the change in the rest of Canada. This model has been used extensively to measure the impacts of health care policies, including the effects of public reporting of hospital performance on mortality [26] and the effects of an intensified diabetes management program [27]. A similar model was also used to measure the effects of mandated insurance coverage for endoscopy screening in the U.S. on CRC screening rates among adults aged 50–64 [28].

The total sample based on province-years that administered the CRC screening module was 81,262, which was reduced to 58,142 after excluding individuals reporting screening due to family history, CRC treatment, or having colitis (10,030) and restricting the data for complete case analysis (13,090). Ontario contributed 19,888 and 11,293 observations and the control provinces 12,324 and 14,637 observations in the pre- and post-intervention periods, respectively.

Comparisons between the intervention and control groups in the pre-intervention period generally show similarities across a range of measures, indicating the appropriateness of the control group. Notable differences include urban geography, ethnic diversity, flu shot uptake, and screening adherence (Table 2). Figure 1 indicates FOBT screening trends in Ontario rose more consistently than those in the control group, and jumped from 15.7% to 21.9% from 2007 to 2008. FOB testing trends in the control group are fairly similar to those in Ontario overall, despite an uncharacteristic decrease in 2005. FOBT use in the control group decreases from 12.4% in 2007 to 10.6% in 2008. Rates of past year endoscopy screening rose gradually in both groups, with little indication of a change in trajectory in 2008.

	Control	Ontario	p		Control	Ontario	p		Control	Ontario	p
				Ethnicity				Leisure physical activity
Socio-demographic				Caucasian	91.5	85.5	0.000	Active	23.5	22.8	0.322
				Other	8.5	14.5		Moderately Active	26.8	26.8	0.995
Age								Inactive	49.6	50.4	0.400
50-64	74.0	73.2		Country of birth
65-74	26.0	26.8	0.159	Canada	81.6	65.4	0.000	Heavy drinking b
								Never	74.4	74.7	0.657
Sex				Time in Canada since immigration				<Once a Month	14.1	13.9	0.654
Male	53.0	52.0	0.136	Non-Immigrant	68.1	61.9	0.000	Once a Month	4.0	3.8	0.601
Female	47.0	48.0		20 or more	26.0	30.7	0.000	2-3 Times a Month	2.6	2.8	0.440
				10 to 19	3.2	4.8	0.001	Once a Week	3.1	2.6	0.058
Marital status				0 to 9	2.7	2.6	0.899	> Once a Week	1.8	2.2	0.058
Married	74.3	73.0	0.067
Common-Law	4.1	4.7	0.082	Health status				Health care use
Widowed	6.6	6.6	0.848
Separated	2.6	2.7	0.560	Self-perceived Health				Has regular MD
Divorced	7.3	7.6	0.395	Poor	3.6	3.7	0.730	Yes	93.3	94.4	0.007
Single, Never Married	5.2	5.4	0.614	Fair	11.7	11.4	0.593
				Good	30.2	29.8	0.592	Screening in past year: FOBT
Highest level of education				Very Good	36.3	35.9	0.581	Yes	8.4	13.5	0.000
< Secondary School	24.9	20.0	0.000	Excellent	18.2	19.3	0.109
Secondary School	17.3	18.2	0.181					Screening in past year: endoscopy
Some Post-Secondary	6.2	6.2	0.966					Yes	3.3	6.0	0.000
Post-Secondary	51.7	55.6	0.000	Body mass index
				Underweight	1.0	1.1	0.442	Ever had a flu shot
Household income a				Overweight	41.0	41.1	0.831	Yes	52.7	70.7	0.000
Quintile 1 (Lowest)	19.9	16.0	0.000	Obese-Class I	15.7	15.5	0.709
Quintile 2	20.0	18.4	0.029	Obese-Class II	3.7	3.5	0.706	No. GP consultations past year
Quintile 3	19.6	19.7	0.902	Obese-Class III	1.5	1.5	0.919	0 to 3	62.3	65.7	0.000
Quintile 4	18.9	20.1	0.080					4 to 10	30.4	28.9	0.062
Quintile 5 (highest)	21.7	25.8	0.000	Health behaviours				11 to 19	5.8	4.4	0.000
								20 or more	1.6	1.0	0.005
Geography				Type of smoker
Urban	69.1	82.8	0.000	Daily	15.5	15.6	0.883	Year
Rural	30.9	17.2		Occasional	2.7	3.0	0.397	2003	60.2	15.3	0.000
				Never	81.8	81.4	0.603	2005	30.2	53.6	0.000
								2007	9.6	31.1	0.000

N=30,484; Weighted N=1,468,619; All proportions weighted. ^aBased on national composition excluding territories. ^bImbibing five or more drinks on one occasion.

We used a quasi-experimental causal approach to measure the effect of ColonCancerCheck on the past-year screening behavior of asymptomatic average risk adults ages 50–74. We found evidence that ColonCancerCheck increased screening adherence within the time frame studied. Past-year screening rates using FOBT increased by an absolute 5.16 [3.19, 7.12] percentage points, equivalent to a 33% increase relative to screening rates in Ontario in 2007. Our findings were robust to different model specifications. Additionally, we failed to find any evidence that past-year endoscopy screening rates were affected by the program. Subsequent tests of model suitability did not indicate evidence of effect modification by any of the factors investigated, or sources of bias. Differences in demographic and medical characteristics between the intervention and control groups were minor and accounted for by the DD model which controls for time-invariant differences across groups. Bias may result if underlying variables were to change alongside the intervention, although inter-group analyses and DDD models did not find any evidence of such interactions.

Although there has been some evidence for increases in FOBT use in Ontario in 2008 [19, 23], this study is the first to estimate the proportion of these increases which can plausibly be attributed to ColonCancerCheck. To our knowledge, this is also the first study to analyze the direct impact of a CRC screening program using a quasi-experimental model. It is unsurprising that a program effect was only observed for FOB testing as the FOBT was the front line test for CRC screening [23] and is recommended to be administered more frequently than other screening tests. On the other hand, screening rates for colorectal cancer are much lower than for other cancers, so a priori it was not obvious whether the program would succeed in overcoming evident barriers to screening and increase screening rates.

In contrast, recommendations for endoscopy in our target group call for its use in screening at up to 10 year intervals [9] and as a follow-up test to positive FOBT results. Length of time to follow-up with endoscopy and low adherence for the procedure may account for our lack of observed effect. Results from the Ontario Pilot recorded median follow-up times from positive FOBT to endoscopy of 121 days in men and 202 in women [40]. Preliminary program results measured follow-up endoscopy within six months following a positive FOBT at 62.1% [23] with only 55% being performed within the target eight weeks [41]. Supply constraints and significant waiting times in the Canadian context [42] are other factors which could lead to delays. The time frame of this study was therefore likely insufficient to capture a potential increase in endoscopy use resulting from increased FOBT screening. Additionally, low follow-up may be explained by the fact that one third of FOB tests performed in Ontario in 2009–10 were outside of the ColonCancerCheck program and did not benefit from the program’s registry and follow-up processes [43]. The restrictions of the study period and the inability to assess program-specific clinical outcomes are inherent constraints of the study, and offer important avenues for future research.

We did not observe evidence of effect modification by having a regular physician for FOBT screening in post-intervention Ontario, which was somewhat surprising given the central role of primary care providers in the program [23]. One possible explanation is that outreach to individuals without PCPs was improved through the program. Secondary analyses [34] found similar increases in screening amongst those with a regular doctor in both Ontario and the other provinces over time, which supports the idea of temporal, rather than program-derived, influence of PCPs on patient screening.

There are several study limitations that should be addressed. First, because the colorectal cancer screening module was not asked in all provinces in all survey waves, we were required to use an unbalanced panel for our control group. This may have disproportionately represented Eastern Canada and biased screening trends in the pre-intervention period downward. This is a concern because, even though the DD and DDD designs can control for time-invariant group differences, alteration of the control group can distort trends and lead to differential results. In particular, the year 2005 indicated a decrease in proportion screened from the previous survey year from 8.34% to 7.03% for the FOBT outcome which was uncharacteristic of the overall trend. However, Table 5 indicates that the DD model is robust to unmeasured trends given the lack of an estimated effect in the pseudo-treatment period.

A second concern is residual confounding based on time-varying unmeasured factors in this observational study. To address this concern, we examined several interactions including having a regular doctor, being aged 65–74 years and 35–49 years, being “inactive”, and reporting ever having had a flu shot. These were chosen given their face validity and common citation in the literature as modifiers of screening. Given no evidence of effect modification by these factors, and no sources of bias found in any other verification test, it seems unlikely that other confounders would bias our estimates appreciably. Although past year outcomes were used to reduce exposure misclassification, the retrospective nature of the reporting likely results in some degree of misclassification and potential attenuation of the treatment effect. Underestimation of standard errors through auto-correlated residuals is common to DD [44], however the relatively short time series and independent nature of the dependent variables mitigates this. Measurement error in self-reported screening rates is another concern, with FOBT recall accuracy estimated to have a sensitivity of 82% and specificity of 78% [45]. Assuming that potential reporting errors are similar in the treatment and control groups given consistency in mode of collection and phrasing of questions [35], we expect the DD design to protect against the effects of these errors.

Finally, there is the possibility of reverse causality, where previous conditions or legislation in Ontario brought about the decision for policy change and independently affected subsequent screening outcomes. We believe that due to federal oversight establishing guarantees for reasonable access to care, harmonized training of healthcare professionals, and conditional transfer payments from the federal government to provinces [17], the concerns of heterogeneity of health services delivery across provinces are dramatically reduced and the likelihood of endogenous factors driving policy change diminished. In Canada, the most severe disparities in health services delivery occur along north–south lines [17] which is in partly why the three Canadian territories were excluded from the analysis. Furthermore, that ColonCancerCheck utilizes previously established national screening guidelines common to all of Canada [7–9], and two other provinces in addition to Ontario are adapting similar programs, further lessens this concern.

Our use of a quasi-experimental approach to derive a causal measure of the impact of the CRC screening program ColonCancerCheck from CCHS microdata files demonstrates the effectiveness of the program in increasing past-year screening by FOBT up to the end of 2009. The results from this study are largely generalizable to other Canadian provinces [46] in addition to other countries [47, 48] which, in many cases, have commonalities in target populations and screening strategies. Insofar as ColonCancerCheck has been successful at increasing rates of screening with FOBT in the asymptomatic average risk population, this study provides evidence that the educational, instructional, and other outreach strategies adopted by the program have been successful. At this point, we are unable to make conclusions as to the effect of the program on use of endoscopy as it is unlikely for an appreciable effect to be observed so soon after program initiation. Another important caveat to this conclusion is that jurisdictions with varying pre-intervention screening levels may experience different degrees of outcomes, so the generalizability of the effect measure should be interpreted with caution. We cannot speculate on the long-term effects of ColonCancerCheck given that it is ongoing [23], that its ability to induce sustained patterns of preventive screening is unknown, and that it must accommodate updated screening recommendations [10]. Updated evidence of the protective effect of flexible sigmoidoscopy [49] and the improved sensitivity of [50] and test adherence to [51] fecal immunochemical tests is encouraging. The conclusions of the current study suggest that similar programs will be capable of promoting these procedures in the average risk population. Among the most important questions in the effectiveness of the program is its ability to reduce morbidity and mortality from CRC. This is not addressed by this study and presents an important avenue for future research examining mortality, costs, stages of cancer at detection and other important outcomes. Given previous evidence of the effectiveness of CRC screening program-like interventions in reducing CRC mortality [11] our results provide some level of optimism that this aim will become fulfilled.