Understanding hepatopancreatobiliary cancer risks in a population-based primary sclerosing cholangitis–inflammatory bowel disease cohort

Abstract

Background and Aims:

Primary sclerosing cholangitis (PSC) is a pre-malignant condition with elevated risk of hepatopancreatobiliary cancers (HPBCa) and colorectal cancer (CRC) when inflammatory bowel disease (IBD) is present. We described cancer burden in a contemporary PSC–IBD cohort, and assessed rates of subsequent cancers, liver transplant, and death post-colectomy or post-cholecystectomy status.

Methods:

Using linked health administrative databases, we calculated cause-specific cumulative incidences of HPBCa-related outcomes (diagnoses/deaths) and non-HPBCa-related transplant/death among individuals with PSC–IBD in Ontario, Canada (2002–2018), followed to 2021. Transition probabilities and transition intensity ratios (TIR) were evaluated using a multistate Markov model.

Results:

Among 476 individuals with incident PSC–IBD, there was a 54% probability of remaining event-free at 10 years. Approximately 1 in 20 experienced an HPBCa-related outcome, and up to 1 in 4 had a non-HPBCa-related transplant/death. During follow-up, 13% experienced multiple events. Age was associated with HPBCa (HR 1.02, 95% CI 1.01–1.04), but not sex. Mortality occurred more frequently post-colectomy (TIR 3.08, 95% CI 1.7–5.59) and post-cholecystectomy (TIR 3.85, 95% CI 2.21–6.64) relative to event-free PSC–IBD, but there were no differences in post-surgery incidence of cancer or transplant.

Conclusions:

While some individuals with PSC–IBD experienced an extended event-free disease course, a large proportion experienced disease-related cancer, colectomy, cholecystectomy, and transplant events. Higher mortality rates observed after surgery were likely related to underlying disease processes that motivated surgical intervention (eg, dysplasia/malignancy, refractory IBD), rather than the surgery itself. Understanding how PSC–IBD disease trajectories vary can inform individual management and patient counselling.

INTRODUCTION

Primary sclerosing cholangitis (PSC) is a chronic, immune-mediated disease causing bile duct inflammation, obstruction, cholestasis, and fibrosis, with >70% with concomitant inflammatory bowel disease (IBD).¹ There is significantly increased cancer risk with PSC–IBD: this includes a 40-fold higher risk of hepatopancreatobiliary cancer (HPBCa),² which comprises cholangiocarcinoma (CCA), hepatocellular carcinoma (HCC), gallbladder carcinoma (gallCa), and pancreatic cancer (pancCa).^3–6 Substantial mortality burden is associated with HPBCa in this population, accounting for 26% of fatalities in PSC patients.⁷ There is also a 7-fold increased risk of colorectal cancer (CRC) with PSC–IBD compared with the general population⁸ and a 3.5-fold increased risk of CRC/dysplasia compared with IBD alone.^3,9,10 Concurrently, PSC is a progressive disease with end-stage complications including cirrhosis, decompensation, or death.¹

Historically, evaluating disease trajectory, cancer risk, and the impact of surgery on outcomes in PSC–IBD populations is challenging, given small sample sizes and referral bias inherent to retrospective cohort studies. As well, risk is often difficult to ascertain in the presence of disease heterogeneity, multiple competing risks, and differential IBD-related exposures (i.e., colectomy, medications). This is evidenced by the ongoing debate in the literature surrounding surgical interventions and associated prognosis in PSC–IBD. As well, individuals with PSC–IBD experience significant anxiety when faced with a diagnosis known for heightened malignancy and mortality, which negatively affects quality of life.¹¹

We sought to describe the burden of disease-associated malignancies (HPBCa and CRC) in a population-based PSC–IBD cohort using health administrative data in Ontario, with a focus on understanding the cumulative incidence of HPBCa compared with non-HPBCa-related morbidity and mortality. We then assessed rates of subsequent HPBCa, CRC, liver transplantation, and all-cause mortality among PSC–IBD individuals following colectomy or cholecystectomy. Lastly, we compared the relative incidence rates of HPBCa and CRC among PSC–IBD compared with IBD alone.

METHODS

The study was approved by the institutional Research Ethics Boards at the University Health Network (20-6302, 22-5209) and Clinical Trials Ontario (ID 3495), and was conducted in accordance with the Declarations of Helsinki and Istanbul. Reporting of methods is in accordance with the REporting of studies Conducted using Observational Routinely-collected health Data Statement (RECORD).¹²

Study design and data sources

We conducted a population-based retrospective cohort study in Ontario, Canada’s most populous province (with ~15 million residents during the study timeframe), from January 1, 2002, to December 31, 2021. The study data resides at ICES, an independent, nonprofit research institute that houses an electronic repository of large, comprehensive population and health administrative datasets for all Ontario residents eligible for the province’s single-payer healthcare system (the Ontario Health Insurance Plan [OHIP]). This covers >99% of the population, with estimated annual provincial out-migration leading to loss to follow-up <0.5%.¹³ These datasets were linked using unique encoded identifiers and analyzed at ICES (see Supplemental Descriptives, http://links.lww.com/HEP/K364). The legal status of ICES under Ontario’s health information privacy law allows it to collect and analyze health care and demographic data, without consent, for health system evaluation and improvement.

Study population

Both PSC–IBD and IBD cohorts have been established at a population level in Ontario using validated health administrative data algorithms.^7,14–16 In brief, a previously validated PSC–IBD algorithm was used for establishment of the incident PSC–IBD cohort in Ontario by: (1) having an IBD diagnosis defined by the Ontario Crohn and Colitis Cohort (OCCC) (1991–2018) (Supplemental Table S1, http://links.lww.com/HEP/K364); and (2) having 1+ PSC-related code (ICD-10-CA code K83.0 cholangitis or K83.00 primary sclerosing cholangitis) in the Discharge Abstract Database (DAD) or National Ambulatory Care Reporting System (NACRS). Incident cases were coded with the first instance of a PSC-related code between January 1, 2002, and December 31, 2018, with a lookback period from April 1, 1988 (using the ICD-9-CA code 576.1 cholangitis). Individuals were excluded if they lacked a unique ICES identifier or had <1 year of OHIP eligibility, if liver transplant or death occurred on or before use of PSC-related codes, or if concomitant diagnostic codes were used for hemochromatosis, alpha-1-antitrypsin deficiency, and/or Wilson disease (Supplemental Table S2, http://links.lww.com/HEP/K364). Diagnostic accuracy of the PSC–IBD identification algorithm in comparison to gold standards had similar performance to commonly used health administrative data algorithms,^17,18 with sensitivity 61%–82%, specificity 100%, and positive predictive value (PPV) 85%–100%.⁵

This study population was a subset from the incident PSC–IBD cohort⁵ where both incident PSC and incident IBD were diagnosed between January 1, 2002, and December 31, 2018 (Figure 1). This population was chosen to (1) evaluate the complete disease trajectory using the available health administrative data in ICES (as many of the ICES databases utilized started after 2000), and (2) to allow for direct comparison to a contemporary incident cohort of individuals with IBD alone during the same timeframe. Index date (time 0) was (1) the earliest of either IBD or PSC diagnosis (PSC/IBD), or (2) PSC diagnosis as specified. The earliest of either IBD or PSC diagnosis was defined as the key index date of interest to reflect the earliest point at which disease pathophysiology was likely active, particularly as most PSC–IBD patients have IBD preceding PSC.⁷ This approach assumes that the order of diagnostic presentation reflects clinical and biological heterogeneity rather than distinct sequential disease processes.¹⁹ Accordingly, individuals presenting initially with IBD may already harbor subclinical PSC that remains undetected due to asymptomatic presentation or normal biochemical parameters.²⁰ Complimentary analyses with PSC date of diagnosis as time 0 are also presented.

FIGURE 1.

Cohort derivation using health administrative data. Abbreviations: IBD, inflammatory bowel disease; PSC, primary sclerosing cholangitis.

A comparator cohort with IBD alone was derived from OCCC, with included individualshaving an incident IBD diagnosis (but no PSC diagnosis) between January 1, 2002, to December 31, 2018, inclusive. Follow-up time for both cohorts was censored at death or December 31, 2021 (end of study period).

Demographics, outcomes, and exposure definitions

Demographic descriptives for the PSC–IBD and IBD alone cohorts were evaluated, including sex and age at PSC and/or IBD diagnoses.

The primary outcomes of interest were PSC–IBD-related cancers, specifically HPBCa (a composite comprising CCA, HCC, gallCa, and/or pancCa), and CRC identified from the Ontario Cancer Registry (OCR) (Supplemental Table S3, http://links.lww.com/HEP/K364), which captures diagnostic information and date of cancer diagnosis for 98% of all incident cases of histologically confirmed cancers in Ontario.^21,22 Underlying cause-of-death was evaluated when available from the Ontario Registrar General Database (ORGD) (Supplemental Table S3, http://links.lww.com/HEP/K364). The time elapsed between PSC diagnosis to HPBCa diagnosis and IBD diagnosis to HPBCa diagnosis was assessed. Liver transplant or death not related to HPBCa (defined by the absence of HPBCa codes in OCR) was considered a competing risk in analyses (Supplemental Table S4, http://links.lww.com/HEP/K364).

The primary exposures of interest were colectomy and cholecystectomy, obtained using OHIP billing claims and procedural codes as used in other ICES health administrative data studies (Supplemental Table S4, http://links.lww.com/HEP/K364). In addition, ileostomy with colectomy codes were also assessed, as these procedures frequently occur concurrently with colectomy (Supplemental Table S4, http://links.lww.com/HEP/K364). As patients may have had certain procedures performed before PSC or IBD diagnoses (ie, colectomy and/or cholecystectomy), procedures done before the index date were also noted, with the earliest available procedure date being April 1, 1988.

Statistical analyses

Median and interquartile ranges (IQR) were used for describing continuous variables, while count and percentage were used for categorical variables. Group differences were assessed using chi-square for categorical variables and Kruskal–Wallis tests for continuous variables. Overall transplant-free survival and HPBCa/transplant-free survival were calculated with the Kaplan–Meier curves.

Cumulative incidence curves for HPBCa-related outcomes in the PSC–IBD cohort were calculated using a cause-specific competing risk approach,²³ adjusting for age and sex. The primary outcome of interest was HPBCa-related outcomes (ie, diagnosis and/or related transplant/death), with non-HPBCa-related transplant/death as the competing risk. Cumulative incidence for both the primary outcome and the competing risk was expressed as a percent probability at 5-year, 10-year, and 15-year intervals for median-aged males and females with PSC–IBD. A sensitivity analysis was conducted using the Fine–Gray subdistribution hazards model adjusted for age and sex.

To investigate rates of subsequent HPBCa, CRC, transplant, or death after colectomy or cholecystectomy relative to event-free PSC–IBD, a time-homogeneous Markov multistate model was applied.²⁴ This model mapped the heterogeneous PSC–IBD disease course by dividing follow-up into predefined states and tracking individuals along a timeline anchored by the index date (earliest of PSC/IBD diagnosis) (Figure 2). Event-free disease was the entry state for most individuals at the index date; individuals with a history of colectomy or cholecystectomy before the index date entered the multistate model as post-colectomy or post-cholecystectomy, respectively. Individuals could progress through states (Figure 2) without re-entering state 1 (event-free PSC–IBD) or leaving state 7 (death). The multistate model was used to estimate (a) transition intensity ratios (TIR) relative to event-free PSC–IBD (conceptually similar to hazard ratios in standard survival analyses), (b) the probability of transitioning to other states at 5 and 10 years, and (c) the “mean sojourn time” (ie, the average time spent in a given state).²⁵ Median sojourn times are not calculated by default in the multistate model due to subjectivity to censoring, skewed data, as well as inability to calculate when >50% have not transitioned out of a particular state during the follow-up period. Given small numbers for some transitions, models were fitted without covariates to examine state changes over time; second or more occurrences of each state were not evaluated separately.

FIGURE 2.

Multistate model representing disease states in individuals with PSC–IBD. Arrows represent possible transitions of individuals between states (Note: transitions where <6 individuals transitioned are not visualized here but are included in the model). Individuals may enter the model as having event-free disease at the beginning of follow-up; a small number of individuals entered the model at the beginning of follow-up status post-cholecystectomy or colectomy. Created with BioRender.com.

Several sensitivity analyses were performed, including evaluating the cumulative incidence of HPBCa outcomes and non-HPBCa-transplant/death after excluding HPBCa diagnoses within the first 2 years of PSC diagnosis, evaluating the impact of colectomy/cholecystectomy as a time-dependent variable on the occurrence of HPBCa-related outcomes (with non-HPBCa transplant/death as a competing risk) with time 0 as PSC diagnosis using Fine–Gray subdistribution hazard models, as well as Cox time-dependent regression analysis to a combined outcome of transplant/all-cause death. Cox regression analysis was also performed from the time of colectomy to assess the role of CRC preceding colectomy and the number of IBD-related hospital presentations in the 2 years before colectomy (as a proxy for IBD-related disease activity/severity) on the risk of post-colectomy death. An additional analysis was also performed to assess HPBCa cumulative incidence from being event-free, experiencing multiple (ie, 2+ major) events.

Crude incidence was calculated for CRC and HPBCa for both the PSC–IBD cohort and the IBD alone cohort. In the case of CRC incidence, contributing person-years (PY) were calculated from index date to total colectomy date or death; similarly, contributing PY were calculated from index date to liver transplant or death for liver-related cancer incidence, and from index date to cholecystectomy or death for gallCa incidence. To contrast the incidence of the various PSC–IBD-related cancers, incidence rate ratios (IRR) were calculated to compare incidence among PSC–IBD to IBD alone.

All analyses were performed using SAS software (version 9.4; SAS Institute Inc., Cary, North Carolina) or R Statistical Software (version 3.1.2; R Foundation for Statistical Computing, Vienna, Austria). All statistical tests were performed at an alpha level of 0.05. Due to Ontario and ICES privacy regulations, the presented small cells are suppressed (n<6), and back-calculation is suppressed by using ranges (eg, 6–10).

RESULTS

Burden of disease-related cancers in PSC–IBD

PSC–IBD cohort descriptives

From the incident PSC–IBD cohort derived from healthcare administrative data in Ontario (n=1076), 476 individuals (59% male) had both PSC and IBD diagnosed between 2002 and 2018 (Figure 1). Median age at PSC diagnosis was 36 y (IQR: 22–54) while median age at IBD diagnosis was 34 y (IQR: 18–52). IBD subtype was ulcerative colitis in 68%, Crohn's in 27%, and undefined for the remainder. Most individuals were diagnosed with IBD first (n=348, 73%); median time from IBD to PSC diagnosis was 3.2 y (IQR: 0.9–6.7). Conversely, 27% were diagnosed with PSC first (n=128, 27%), with a median time from PSC to IBD diagnosis of 0.9 y (IQR: 0.06–3.9). Median follow-up time from the earliest of PSC/IBD diagnosis date to death/censor date for the PSC–IBD incident cohort was 12.0 y (IQR: 8.5–15.7) (IBD then PSC, 11.8 y [8.0–15.3]; PSC then IBD, 12.9 y [9.5–16.4]). Area-level sociodemographic details for the cohort has been previously described⁷ (also available in Supplemental Table S5, http://links.lww.com/HEP/K364).

Median transplant-free survival from the earliest of PSC/IBD diagnosis was 11.1 y (IQR 7.6–14.7); from PSC diagnosis, median transplant-free survival was 7.1 y (IQR 3.9–11.4). Median age at transplant and all-cause death was 36 y (IQR 23–44) and 63 y (IQR 47–80), respectively. The type of donor was ascertained in n=67, of which 60% were deceased donor transplants and 40% were live donor transplants. Traditional Kaplan–Meier derived 5-year and 10-year probabilities of transplant-free survival from PSC/IBD diagnosis were 89% (95% CI 87%–92%) and 76% (95% CI 72%–80%), respectively. The equivalent for HPBCa/transplant-free survival were 89% (95% CI 86%–92%) and 75% (95% CI 71%–79%), respectively.

HPBCa and CRC are most often diagnosed among middle-aged PSC–IBD individuals

There were 36 individuals diagnosed with HPBCa in the PSC–IBD cohort during the study period, with 20 (4.2%) individuals diagnosed with CCA (6 intrahepatic only, 14 with extrahepatic involvement [including mixed, ampullary, and undetermined CCA]), 7 (1.5%) with HCC, 6–10 with pancCa, and <6 with gallCa. Among the PSC–IBD cohort, the median time from PSC diagnosis to HPBCa diagnosis was 0.3 y (IQR: 0–4.7 y), while the median time from IBD diagnosis to HPBCa diagnosis was 7.3 y (IQR 4.3–10.7) (Supplemental Figure S1, http://links.lww.com/HEP/K364). Median age at diagnosis with CCA was 47 y (IQR 31–65; intrahepatic-only CCA: 44 y, IQR 26–65; CCA with extrahepatic involvement: 50 y, IQR 43–65). Median age at HCC diagnosis was 52 y (24–61), pancCa was diagnosed at a median age of 57 y (IQR: 40–80), and gallCa was diagnosed at a median age of 62 y(IQR 57–84). There were 13 individuals (2.7%) diagnosed with CRC among those with PSC–IBD; median age at CRC diagnosis was 46 y (IQR: 41–67).

Higher cumulative incidence of non-HPBCa-related transplant/death is seen compared with HPBCa-related events

Non-HPBCa-related transplants/deaths were more frequent (n=126, 26%) than HPBCa-related outcomes (n=38, 8.0%; 36 diagnosed during follow-up, remainder attributed only at death). Using a competing risk approach with cause-specific analyses, age at index date (ie, diagnosis of PSC/IBD) was significantly associated with HPBCa risk (HR 1.03 per 1 y increase, 95% CI 1.01–1.04), while sex was not associated (HR 1.54, 95% CI 0.78–3.06, reference group: females). Similarly, Fine–Gray subdistribution hazards modeling substantiated that increasing age at index date was associated with increased HPBCa risk (HR 1.02 per 1 y increase, 95% CI 1.01–1.04), but not sex (HR 1.44, 95% CI 0.73–2.85). The cumulative risk of HPBCa-related outcomes was much lower compared with the cumulative risk of non-HPBCa-related transplant/death, as demonstrated in Table 1 at 5, 10, and 15 years. Measured from the PSC diagnosis date, the cumulative risk of HPBCa-related outcomes was slightly higher at 5 years for both sexes, in keeping with cancer diagnoses occurring shortly after PSC diagnosis. Similarly, non-HPBCa-related transplant/death cumulative risks were higher when measured from the PSC diagnosis date. Sensitivity analyses with cumulative risk calculations after excluding individuals with HPBCa diagnosed within the first 2 years of PSC diagnosis demonstrated reduced cumulative risk of HPBCa-related outcomes, but higher cumulative risk of non-HPBCa-related events (Supplemental Table S8, http://links.lww.com/HEP/K364).

TABLE 1.

Cumulative risk of hepatopancreatobiliary cancer (HPBCa)-related outcomes and non-HPBCa-transplant/death using cause-specific competing risk regression analyses

From the earliest PSC/IBD diagnosis	5 y	10 y	15 y
HPBCa-related outcomes, % (95% CI)
Malea	1.4 (0.5–3.8)	5.0 (2.9–8.6)	7.9 (4.9–12.6)
Femaleb	0.5 (0.1–2.2)	3.1 (1.4–6.6)	6.2 (3.3–11.3)
Non-HPBCa-related transplant/death, % (95% CI)
Malea	7.5 (5.0–11.4)	18.2 (14.0–23.6)	27.8 (22.3–34.6)
Femaleb	7.9 (4.9–12.8)	15.9 (11.4–22.2)	24.1 (18.2–31.9)
From the PSC diagnosis	5 y	10 y	15 y
HPBCa-related outcomes, % (95% CI)
Malea	3.5 (1.9–6.5)	4.7 (2.5–8.7)	6.2 (3.7–10.3)
Femaleb	3.5 (1.9–6.5)	5.3 (3.1–9.0)	6.7 (3.3–13.5)
Non-HPBCa-related transplant/death, % (95% CI)
Malea	15.4 (11.5–20.6)	26.5 (21.8–32.1)	33.1 (26.6–41.0)
Femaleb	13.0 (9.1–18.7)	20.4 (15.6–26.7)	28.2 (20.9–38.0)

^a Calculated for the median-aged male and female with primary sclerosing cholangitis–inflammatory bowel disease (PSC–IBD). Median age for males: 31 y.

^b Median age for females: 36 y. Note that there was no statistically significant difference between sexes for either outcome.

Modeling state transitions, state probabilities, and transition intensity ratios in PSC–IBD

Disease trajectory in PSC–IBD is diverse and complex

Among the 476 individuals included, there were 117 (25%) individuals with cholecystectomies, 109 (23%) with colectomies, 98 (21%) with deaths, 80 (17%) who underwent liver transplant, 36 (7.6%) with histologically proven HPBCa, and 13 (2.7%) with histologically proven CRC. A diagnosis of HPBCa was associated with worse survival than other disease states (Supplemental Figure S3, http://links.lww.com/HEP/K364).

There were 47 different disease transition trajectories during follow-up: the majority were stable in 1 state (n=214, 45%) or only transitioned through 2 states (n=173, 36.3%), while the remainder transitioned through 3 to 5 states (3: n=62, 13%; 4: n=21, 4.4%; 5, n=6, 1.3%) (Figure 2 and Supplemental Table S6, http://links.lww.com/HEP/K364). Remaining event-free from index date to the end of the follow-up period was the most common trajectory observed (n=190, 40%); the next most common trajectories observed for those starting event-free (n=425) involved a single transition from event-free PSC/IBD to colectomy (n=47, 11%), to transplant (n=41, 9.6%), to cholecystectomy (n=30, 7.1%), and to death (n=30, 7.1%). There were 14 individuals who transitioned from event-free status to HPBCa to death (3.3%). The 5-year and 10-year probabilities of experiencing multiple (ie, 2+) major events from the event-free state were 4.3% (95% CI 2.3–6.2) and 13.3% (95% CI 9.7–16.7), respectively.

Of individuals who had a cholecystectomy at or before PSC–IBD diagnosis (n=40), most (n=18, 45%) did not transition to other states, while 8 (20%) and 9 (23%) transitioned directly to colectomy and death, respectively. There were 59 individuals who had a cholecystectomy during follow-up, for which approximately half (31/59, 53%) did not transition to another state for the remainder of follow-up; there were <6 individuals diagnosed with HPBCa before cholecystectomy, 9 individuals who had HPBCa diagnosed and cholecystectomy concurrently, and as well than <6 diagnosed with HPBCa after cholecystectomy. Of all individuals who had a cholecystectomy at any point, 42/117 (36%) experienced transplant and/or death during follow-up, as opposed to 75 (64%) who did not. There were 31 deaths among individuals with prior cholecystectomy, with the most common causes of death listed in Supplemental Table S7C, http://links.lww.com/HEP/K364.

Of individuals who had a colectomy at or before PSC–IBD diagnosis (n=11), there was a median of 1.07 y (IQR 0–6.2 ) between colectomy and IBD diagnosis; 6 (55%) did not transition to other states. Of individuals who had a colectomy during follow-up (n=82), 58 (71%) did not transition to other states for the remainder of follow-up, while <6 had a diagnosis of CRC preceding colectomy, 9 had concurrent CRC diagnosis and colectomy, and <6 had a diagnosis of CRC following colectomy (presumed due to the colectomy being only partial). Of all individuals who had a colectomy at any point, 27/109 (25%) experienced transplant and/or death during follow-up, in contrast to 82 (75%) who did not. There were 21 deaths post-colectomy with a median time elapsed between colectomy and death of 4.5 y (IQR 1.1–9.9). Most common underlying causes of death among individuals who had prior colectomy are listed in Supplemental Table S7D, http://links.lww.com/HEP/K364. The time to death from colectomy preceded by CRC was shorter (median 1.1 y, IQR 1.1–2.4) than death post-colectomy without preceding CRC (median 5.1 y, IQR 1.4–12.7); however, progressing from colectomy to death was not significantly associated with CRC preceding colectomy (HR 2.86, 95% CI 0.64–12.7). As well, IBD-related presentations to hospital (0, 1, 2+ in the 2 y prior) before colectomy were also not predictive of death (HR 0.95, 95% CI 0.76–1.18).

One-third of deaths were preceded by HPBCa during follow-up (n=30, 31%), with 17/30 (57%) individuals transitioning directly from HPBCa diagnosis to death without passing through other states (eg, surgery); others transitioned through other states, including cholecystectomy, colectomy, and/or transplant. Transplant was preceded by an HPBCa or CRC diagnosis in 8/80 (10%) individuals. There were 14/80 (18%) individuals who underwent liver transplant who experienced subsequent death during follow-up, of which the majority transitioned directly (n=11). There were <6 who had been diagnosed previously with HPBCa or CRC before transplant, and <6 who underwent other surgeries and/or were diagnosed with HPBCa after transplant before death.

Individuals who are event-free, post-colectomy, post-cholecystectomy, or post-transplant are usually stable for years

The “mean sojourn time” for a PSC–IBD individual remaining event-free was 16.2 y (95% CI 14.2–18.4). Post-colectomy individuals did not transition for a mean of 22.5 y (95% CI 15.8–32.0), while post-cholecystectomy individuals did not transition for a mean of 14.0 y(95% CI 10.7–18.4). Post-transplant individuals were also stable for a mean of 20.6 y (95% CI 13.3–31.9). The “mean sojourn times” for individuals spent in CRC and HPBCa states before transitioning were 1.60 y (95% CI 0.89–2.89) and 0.82 y (95% CI 0.59–1.14), respectively.

Prediction from different disease states estimates PSC–IBD status at 5 and 10 years

Figure 3A presents the 5-year and 10-year state occupancy probabilities estimated by the multistate model for individuals who started in the event-free PSC–IBD state at the index date, reflecting the likelihood of transitioning to other disease states over time. Remaining event-free at 5 years and 10 years was 73% and 54%, respectively. The probability of death occurring at 5 years and 10 years was 6.3% and 14%, respectively, while the probability of transplant was 6.1% and 10%, respectively. As HPBCa and CRC states were transient compared with the other states, there was a very low likelihood for individuals to be in these states at 5 years or 10 years. Colectomy and cholecystectomy at 5 years were 7.7% and 5.6%, respectively; at 10 years, they were 13% and 8.3%, respectively.

FIGURE 3.

Probability of transitioning to other states from (A) event-free PSC–IBD in 5 years and 10 years, (B) from alternate states in 5 years, and (C) from alternate states in 10 years. Abbreviations: HPB, hepatopancreatobiliary; IBD, inflammatory bowel disease; PSC, primary sclerosing cholangitis. Created with BioRender.com.

For individuals who transition to other states (ie, post-surgery or with a cancer diagnosis), Figures 3B, C demonstrate the transition probabilities to other disease states at 5 years and 10 years, respectively. Post-cholecystectomy and post-colectomy status were stable states, with 70% and 81% remaining, respectively, at 5 years, and 50% and 66%, respectively, at 10 years. Mortality post-procedure was high at 5 years, with 14% of post-cholecystectomy and 12% of post-colectomy individuals dying; by 10 years, 26% of post-cholecystectomy and 23% of post-colectomy individuals were deceased.

Both cancer types carried a significant mortality risk. Twenty percent of CRC and 54% with HPBCa had died by 5 years (Figure 3B). However, liver transplant was still performed in these groups, with 4.6% with CRC and 21% with HPBCa having had a transplant by 5 years. Once CRC was diagnosed, the majority underwent colectomy (68% probability of being status post-colectomy at 5 years). Many individuals also transitioned from HPBCa to cholecystectomy at 5 years (19%). By 10 years from CRC or HPBCa diagnosis, 30% and 61% were deceased.

Post-transplant, most individuals remained stable, with 81% remaining in the post-transplant state at 5 years and 66% at 10 years. However, the probability of being deceased at 5 years and 10 years was 13% and 24%, respectively. Death occurred after transplant in 15 individuals at a median of 43 years (IQR 27–51), with the main cause of death post-transplant listed in Supplemental Table S7B, http://links.lww.com/HEP/K364. The probability of having had a colectomy post-transplant at 5 years and 10 years was 6.9% and 11%, respectively.

Colectomy and cholecystectomy are associated with a higher risk of death among PSC–IBD, but not transplant or HPBCa

Figure 4 demonstrates the TIR of direct progression from the states of post-colectomy, post-cholecystectomy, and HPBCa diagnosis to the main outcomes of interest (HPBCa, CRC, transplant, and/or death) relative to individuals with event-free PSC–IBD. Post-colectomy status was not associated with increased rate of developing future HPBCa (TIR 0.36, 95% CI 0.09–1.42) nor rate of liver transplant (TIR 0.87, 95% CI 0.41–1.84), but a significantly higher rate of death (TIR 3.08, 95% CI 1.70–5.59). Cholecystectomy status was not associated with future direct progression to HPBCa or CRC (Figure 4B). There was no significant association between cholecystectomy status and liver transplantation (TIR 1.46, 95% CI 0.81–2.65); however, there was a significantly increased rate of death after cholecystectomy compared with event-free PSC–IBD (TIR 3.85, 95% CI 2.21–6.64).

FIGURE 4.

Transition intensity ratios (TIR) from (A) colectomy, (B) cholecystectomy, and (C) HPB cancer status to other states, relative to event-free PSC–IBD. Abbreviations: HPB, hepatopancreatobiliary; IBD, inflammatory bowel disease; PSC, primary sclerosing cholangitis. Created with BioRender.com.

A diagnosis of HPBCa was associated with higher progression rates to liver transplant (TIR 20.9, 95% CI 10.0–44.7) and mortality (TIR 75.0, 95% CI 41.4–136.0) (Figure 4C). There was no substantive difference in progression rate to HPBCa from a CRC diagnosis relative to event-free PSC–IBD; similarly, there was no significantly higher progression rate to death (TIR 7.31, 95% CI 0.98–52.6 for both). Transition from CRC to HPBCa was not significant (TIR 7.31, 95% CI 0.98–42.6).

A sensitivity analysis using Fine–Gray subdistribution hazards modeling using PSC diagnosis as time 0 and with cholecystectomy as a time-dependent variable found that cholecystectomy was neither associated with HPBCa development (HR 1.70, 95% CI 0.78–3.73) nor non-HPBCa-related transplant/death (HR 1.87, 95% CI 0.87–4.03). Similarly, sensitivity analyses from PSC diagnosis with colectomy as a time-dependent variable demonstrated that colectomy was not significantly associated with either HPBCa (HR 0.56, 95% CI 0.22–1.50) nor with transplant/death as a competing risk (HR 0.54, 95% CI 0.17–1.77), nor with colectomy as a time-dependent variable with transplant/all-cause death as the outcome of interest (without competing risks) (HR 1.08, 95% CI 0.69–1.68).

Contrasting the burden of HPBCa and CRC in PSC–IBD to the burden in those with IBD alone

IBD comparator cohort descriptives

The background comparator group consisted of 54,591 individuals (49% male) diagnosed with IBD alone, with a median age at IBD diagnosis of 38 years (IQR: 24–54 y). This was significantly later compared with age at IBD diagnosis in those with PSC–IBD (p<0.001). Compared with the PSC–IBD cohort, there was no significant difference in the median follow-up time for the IBD alone incident cohort (11.5 y vs. 10.9 y, p=0.15).

Individuals with PSC–IBD are significantly more likely to be diagnosed with HPBCa and CRC than those with IBD alone

HPBCa and CRC incidence were higher in PSC–IBD than in IBD alone, regardless of whether the index date was the date of PSC diagnosis or the earliest of the PSC or IBD diagnosis date (Table 2). The incidence of liver-related malignancy, HCC, and CCA was significantly higher (~30–60 times) in the PSC–IBD population compared with the IBD alone population. Similarly, CRC among PSC–IBD was relatively elevated (IRR 2.08, 95% CI 0.86–5.02 from the earliest of PSC/IBD diagnosis). There was also a higher risk of gallCa and pancCa among PSC–IBD compared with IBD alone.

TABLE 2.

Incidence and relative risk of common PSC–IBD-related malignancies among PSC–IBD individuals compared to those with IBD alone

	Incidence per 1000 PY (95% CI)
	PSC–IBD (n=476)		IBD alone (n=54,591)	Incidence rate ratios (IRR) comparing PSC–IBD to IBD alone
Cancer type	From the earliest PSC or IBD diagnosis	From the PSC diagnosis	From the IBD diagnosis	From the earliest PSC or IBD diagnosis	From the PSC diagnosis
Livera	6.55 (4.54–9.16)	6.76 (4.38–9.98)	0.14 (0.11–0.17)	48.3 (30.8–75.7)	46.8 (31.3–69.9)
HCC	2.11 (1.05–3.77)	2.70 (1.29–4.96)	0.07 (0.05–0.10)	30.1 (12.8–71.0)	38.6 (19.3–77.0)
CCA	4.41 (2.80–6.62)	4.03 (2.26–6.65)	0.07 (0.05–0.09)	63.0 (32.9–120.7)	57.6 (31.7–104.4)
GallCa	0.57 (0.12–1.68)	0.54 (0.06–1.94)	0.03 (0.02–0.05)	19.0 (5.54–65.2)	18.0 (4.14–78.3)
PancCa	0.96 (0.31–2.23)	0.81 (0.17–2.35)	0.14 (0.11–0.18)	6.86 (2.78–16.9)	5.79 (1.83–18.3)
CRC	2.32 (1.20–4.05)	1.35 (0.44–3.14)	0.65 (0.59–0.72)	3.57 (2.01–6.34)	2.08 (0.86–5.02)

^a Includes HCC, CCA, and ampullary cancers.

Abbreviations: CCA, cholangiocarcinoma; CI, confidence intervals; CRC, colorectal cancer; GallCa, gallbladder cancer; HCC, hepatocellular carcinoma; IBD, inflammatory bowel disease; PancCa, pancreatic cancer; PSC, primary sclerosing cholangitis; PY, person-years.

DISCUSSION

In a population-level PSC–IBD cohort, the cumulative incidence of HPBCa approximated 1 in 20 at 10 years, while accounting for competing risks of non-HPBCa transplant/death (which occurred up to 1 in 4 at 10 y). Multistate modeling demonstrated variable trajectories: many remained event-free throughout follow-up (54% at 10 y), while others transitioned to other states (13% who experienced 2+ events). The model confirmed the high morbidity and mortality of HPBCa, whereas CRC cases largely transitioned to colectomy. Post-colectomy and post-cholecystectomy states carried a 3- to 4-fold higher rate of death compared with event-free PSC–IBD, though neither surgery was linked to elevated future rates of cancer or transplant. PSC–IBD patients had approximately 300-fold higher HPBCa and 4-fold higher CRC rates than IBD alone, underscoring their comparative substantial cancer burden.

In keeping with existing literature,^3,26 HPBCa was diagnosed temporally closer to PSC diagnosis than IBD diagnosis. The temporal proximity of PSC and HPBCa likely reflects delayed PSC diagnosis, as the disease may precede symptoms by years. Because PSC is often underrecognized, subclinical biliary disease with malignant potential may long predate clinical detection. This may explain why IBD duration, possibly reflecting longstanding undiagnosed PSC, is associated with CCA risk.²⁷ Although surveillance bias after surgery is possible, current guidelines already mandate frequent imaging for HPBCa and annual colonoscopy for CRC;^28,29 consistent with this, we observed no accelerated development of HPBCa following colectomy or cholecystectomy.

Identifying individuals with PSC–IBD who are at greatest risk for HPBCa or CRC remains a key clinical objective; however, early detection of HPBCa continues to pose significant challenges. Although male sex, advanced age, and longer IBD duration have been associated with increased CCA risk, substantial uncertainty in risk estimates persists.^27,30 Frequent imaging and regularly scheduled endoscopic retrograde cholangiopancreatography in PSC–IBD may improve transplant-free survival,³¹ although reliable clinical and radiologic markers of early-stage disease are often lacking, and no differences have been found in HPB malignancy development.^32,33 Stringent surveillance is currently recommended,³⁴ although the interval and modality of surveillance remain variable.³¹ Reassuringly, our findings show that progression to HPBCa or HPBCa-related transplant/death is less common than non-HPBCa-related transplant/death. The risk of HPBCa and CRC is a major concern for individuals with PSC–IBD; in one survey, 26% believed their 1-year CCA risk exceeded 20% and 33% estimated a lifetime CRC risk of ≥40%, whereas 19% perceived no risk of CCA within the year.¹¹ The multistate model captures the heterogeneous PSC–IBD disease course by accommodating multiple events over time and segmenting trajectories to provide a clearer picture of progression and individual risk than traditional survival models.

Both cholecystectomy and colectomy were associated with higher subsequent mortality compared with event-free PSC–IBD. Death post-colectomy with preceding CRC occurred earlier than deaths post-colectomy without preceding CRC on sensitivity analyses; however, neither CRC nor IBD-related hospitalizations preceding colectomy were statistically significantly associated with death. It is likely that prognosis post-colectomy reflects the underlying surgical indication—malignancy risk (eg, dysplasia, gallbladder polyps, cancer) or severe disease (eg, refractory colitis, perforation)—rather than the surgery itself, making these procedures markers of advanced PSC–IBD necessitating such surgeries and explaining the elevated mortality. This is supported by evidence that cholecystectomy for gallbladder polyps or masses carries high morbidity in PSC,³⁵ and is associated with greater liver enzyme elevations and more severe biliary strictures and dilatations in those with prior cholecystectomy.³⁶ One study found that colectomy before PSC diagnosis portended improved transplant-free survival, but not colectomy after PSC diagnosis;³⁷ such results may be due to differential indications for colectomy (eg, IBD activity as opposed to malignancy concern).

Limitations include misclassification bias from administrative data, as asymptomatic PSC–IBD is often only coded once clinically overt, biasing findings toward advanced disease and overestimating complication risk. We minimized this using validated diagnostic and procedural coding where able. Data on surgical indications beyond histology-confirmed cancer, non-cancer pathology, dysplasia, IBD severity, medication use, symptom burden, and gallbladder polyp radiology were unavailable for colectomies and cholecystectomies. The surgical indication is likely causally related to the altered mortality rate seen post-surgery, rather than the surgery itself (with surgery serving as a mediator in the causal pathway). Larger multicenter cohort studies with detailed patient-level surgical data are needed to clarify the relationship between specific indications and the outcomes observed, using inverse probability weighting and time-dependent modeling to address confounding. Furthermore, as the cohort required incident diagnoses (2002–2018), median follow-up was ~10 y (max. 20 y) and favored younger patients, especially those with IBD before PSC. Of note, a high mortality rate was seen both pre-transplant and post-transplant. This is likely due to a cohort effect (or selection bias), as our study was limited to PSC–IBD individuals who were diagnosed during a period of 2 decades. Those who required transplant during the timeframe may have had more aggressive disease necessitating transplant, hence contributing to the higher pre-transplant and post-transplant mortality seen in this cohort. As well, the time-homogeneous Markov multistate model assumes constant transition risks over time (time homogeneity) and that future transitions depend only on the current state, not prior history (Markov property). Consequently, this framework may oversimplify the complex and heterogeneous disease dynamics underlying PSC. Due to some transitions in this cohort rarely occurring, sex, age, and other covariates were not tested in the multistate model due to the inability to achieve model convergence. In future research, larger cohorts with a greater number of observed transitions may allow the implementation of a clock-reset multistate model, enabling more nuanced transitions-specific evaluation of age, sex, disease stage, and state-specific duration.

In summary, this population-based study confirms the substantial burden of malignancy-related morbidity and mortality among patients with PSC–IBD, contextualized with the burden of non-malignancy-related morbidity and mortality. Individuals with status post-colectomy or cholecystectomy had accelerated mortality compared with patients who were event-free, likely related to the underlying indication necessitating these surgeries. By accounting for competing risks and estimating progression to HPBCa, CRC, transplant, and death from various PSC–IBD states, the multistate model refines disease trajectory assessment from each state. These findings can inform personalized patient counseling regarding risk and prognosis.

Supplementary Material

hep-84-658-s001.docx ^{(528KB, docx)}