Increasing Prevalence of Primary Biliary Cholangitis and Reduced Mortality With Treatment

Transcription

1 Clinical Gastroenterology and Hepatology 2018;16: Increasing Prevalence of Primary Biliary Cholangitis and Reduced Mortality With Treatment Mei Lu,* Yueren Zhou,* Irina V. Haller, Robert J. Romanelli, Jeffrey J. VanWormer, k Carla V. Rodriguez, Heather Anderson, # Joseph A. Boscarino,** Mark A. Schmidt, Yihe G. Daida, Amandeep Sahota, kk Jennifer Vincent, Christopher L. Bowlus, ## Keith Lindor,*** Talan Zhang,* Sheri Trudeau,* Jia Li,* Loralee B. Rupp, and Stuart C. Gordon, for the Fibrotic Liver Disease Consortium Investigators *Department of Public Health Sciences, Center for Health Policy and Health Services Research, Department of Gastroenterology and Hepatology, Henry Ford Health System, Detroit, Michigan; Essentia Institute of Rural Health, Essentia Health, Duluth, Minnesota; Palo Alto Medical Foundation Research Institute, Palo Alto, California; Marshfield Clinic Research Foundation, Marshfield, Wisconsin; Center for Health Research, Kaiser Permanente Mid-Atlantic Research Institute, Rockville, Maryland; # Skaggs School of Pharmacy and Pharmaceutical Sciences, University of Colorado, Aurora, Colorado; **Department of Epidemiology and Health Services Research, Geisinger Clinic, Danville, Pennsylvania; Center for Health Research, Kaiser Permanente Northwest, Portland, Oregon; Center for Health Research Hawai i, Kaiser Permanente, Honolulu, Hawaii; Department of Research and Evaluation, Kaiser Permanente Southern California, Los Angeles, California; Baylor, Scott and White Research Institute, Temple, Texas; ## University of California Davis School of Medicine, Sacramento, California; ***College of Health Solutions, Arizona State University, Phoenix, Arizona BACKGROUND & AIMS: METHODS: RESULTS: There are few data from longitudinal studies of trends in primary biliary cholangitis (PBC) among patients under routine clinical care in the United States. We collected data from the Fibrotic Liver Disease consortium to investigate changes in the incidence and prevalence of PBC and the effects of patient demographics, clinical features, and treatment on mortality. We collected demographic and clinical data for the general patient population as well as PBC patients receiving care from 11 health systems in different regions of the United States (Northeast, Midwest, Northwest, and South) from January 1, 2003, through December 31, Annual percentage changes in PBC prevalence and incidence were estimated using join-point Poisson regression. Differences based on race, age, and gender were calculated with rate ratios. All-cause mortality was estimated using Cox regression with adjustment for patient characteristics and treatment with ursodeoxycholic acid (UDCA). Propensity scores were used to adjust for treatment selection bias. Analyses were adjusted by geographic regions. In our racially diverse cohort of 3488 patients with PBC (21% Hispanic, 8% African American, 7% Asian American), 70% had ever received UDCA. From 2006 through 2014, the prevalence of PBC increased from 21.7 to 39.2 per 100,000 persons. Adjusted annual percentage changes in prevalence differed among age groups ( 40 y, y, y, y, and >70 y), ranging from 3.0% to 7.5% (P <.05). Incidence did not change significantly during the study period (4.2 vs 4.3 per 100,000 person-years in 2006 and 2014, respectively; P [.98). Ratios of prevalence for women vs men (3.9:1) and incidence for women vs men (3.2:1) were consistent over the study period. Among African Americans, the prevalence of PBC increased from 16.9 to 30.8 per 100,000 during the study period, and annual incidence ranged from 2.6 to 6.6 per 100,000 person-years. In adjusted analyses, an increased level of alkaline phosphatase at baseline was associated with significantly higher mortality (adjusted hazard ratios [ahr], 1.24; 95% CI, for patients with levels 1 2 times the upper limit of normal and ahr, 2.27; 95% CI, for patients with levels more than 3 times the upper limit of normal). UDCA treatment was associated with significantly reduced mortality (ahr, 0.57; 95% CI, ). Abbreviations used in this paper: aapc, adjusted annual percentage change; ahr, adjusted hazard ratio; ALP, alkaline phosphatase; ALT, alanine aminotransferase; AMA, antimitochondrial antibody; APC, annual percentage change; ASINPI, Asian American, American Indian, or Pacific Islander; AST, aspartate aminotransferase; EHR, electronic health record; FOLD, Fibrotic Liver Disease Consortium; PBC, primary biliary cholangitis; RR, rate ratio; UDCA, ursodeoxycholic acid. Most current article 2018 by the AGA Institute /$

2 August 2018 Reduced PBC Mortality With Treatment 1343 CONCLUSIONS: In an analysis of data from patients receiving routine clinical care in Fibrotic Liver Disease Consortium health systems, we found that the prevalence of PBC increased from 2004 through 2014, despite steady incidence. Patient demographic and clinical characteristics, as well as UDCA treatment, affected mortality. Keywords: FOLD Consortium; Epidemiology; Racial Disparities; Gender; Autoimmune Disease. See related article on page There are currently no longitudinal studies of trends in primary biliary cholangitis (PBC) epidemiology among patients under routine clinical care in the United States. The Fibrotic Liver Disease (FOLD) Consortium recently reported an overall 12- year period prevalence of PBC of 29.3 per 100,000 persons 1 ; this prevalence varied significantly by US census region and by patient demographics. Although there are few studies of PBC trends in the United States, studies from Europe and Asia reported a generally steady to slightly increasing PBC prevalence in the past decade, with roughly 4 to 9 times higher prevalence among womencomparedwithmen. 2,3 However, because these studies used data drawn from national registry databases with varying case ascertainment and data collection methodologies, results have not been consistent, and their generalizability to the US population is unknown. Lifelong treatment with ursodeoxycholic acid (UDCA) is recommended for all PBC patients. UDCA has been shown to improve clinical outcomes, particularly when treatment commences in the early stages of the disease before the development of cirrhosis. A recent review of data derived from clinical trials found no benefit of UDCA treatment on mortality, 4 whereas a large US placebocontrolled study with 3 years of follow-up evaluation showed that UDCA treatment improved survival. 5 A recent major editorial 6 emphasized the value of real world evidence regarding drug efficacy, particularly with regard to optimum dosages, patient subgroups, and long-term outcomes, because comorbidities and treatment adherence may differ substantially in the real world compared with clinical trials. However, there are little data regarding the impact of PBC treatment on mortality among patients under routine clinical care in the United States. The FOLD Consortium was established to address this knowledge gap in PBC epidemiology, natural history, and treatment among patients under routine care in the United States. In the present analysis, we used data from the FOLD cohort to investigate the following: (1) changes in incidence and prevalence of PBC over time, (2) the influence of patient demographics on those trends, and (3) the impact of patients clinical characteristics and receipt of UDCA treatment on all-cause mortality. Methods The FOLD Consortium has been described previously. 1 Briefly, FOLD comprises 11 geographically diverse health systems, representing 4 US Census Bureau defined regions of the United States (Northeast, Midwest, Northwest, and South). All sites are members of the Health Care Systems Research Network, sharing a common data structure that permits efficient sharing of electronic health record (EHR)-based data. FOLD follows the guidelines of the US Department of Health and Human Services for the protection of human subjects. The study protocol was approved by the Institutional Review Board of each participating site. All authors had access to the study results, and reviewed and approved the final manuscript. We previously developed and validated an automated Classification and Regression Tree method 7 for accurately identifying PBC patients using EHR data. 1 Classifier variables included the following: (1) an antimitochondrial antibody (AMA)-positive test with at least 1 PBC diagnosis code; (2) an AMA-positive test with a recorded alkaline phosphatase (ALP) level >150 IU/L (in the absence of a PBC diagnosis code); or (3) the presence of 2 PBC diagnosis codes (in the absence of an AMA-positive result). This classification model was validated using 8 independent FOLD data sets, and had an area under the receiver operator characteristic curve value of more than 93% (which is considered excellent classification accuracy). Data Collection and Trend Analysis for Trends: Patient demographic data age category at index date (40, 41 50, 51 60, 61 70, and >70 y), race (black/african American, Asian American/American Indian/Pacific Islander [ASINPI], white, and other/ unknown), and region (Midwest, South, West, and Northeast) were collected for both PBC cases and the overall health system populations. Retrospective data were collected from January 1, 2003, through December 31, Calculations of prevalence and incidence were begun 3 years into the study period (from 2006 onward) to allow a wash out period, following an approach proposed by Lleo et al. 3 Patients diagnosed before 2006 were captured in PBC prevalence estimates for later years if they remained in a FOLD health system. The index date was defined as the earliest date of either PBC

3 1344 Lu et al Clinical Gastroenterology and Hepatology Vol. 16, No. 8 diagnosis or AMA-positive test result during the study period. Prevalence for a given year included all living PBC cases with an index date through December 31 of that given year. Incidence for a given year was calculated based on the number of PBC cases whose index date fell into a given year. Individual patients were included in the denominator for all years in which they had an encounter with a FOLD health system, as well as any years between their first and last encounter regardless of whether they had a health system encounter. Finally, we plotted the observed overall rates of PBC prevalence and incidence by time, as well as those rates by age, gender, and race. Join-point modeling was used to study the dynamics of longitudinal trends in prevalence and incidence. We adapted and extended a 2-step join-point Poisson regression modeling approach 8 by fitting a series of straight lines on a log scale to the trend; each join-point represents a statistically significant (P <.05) change in trend (ie, slope of the line segment). For example, a single join-point splits the trend line into 2 line segments, whereas zero join-points indicates that the best fit to the trend consists of only a single line segment. In the first step, we identified the optimal join-point(s) using a nonlinear modeling approach. Next, multivariable analyses were performed based on the selected join-point(s) as well as potential stratification variables. Interactions were tested only for individual variables that were significant. Variables were retained in the final model if the estimated annual percentage changes (APCs) before and after the join-point were significant (P <.05). We also evaluated whether the APC of each line segment differed from no change (APC ¼ 0). Variables were retained in the adjusted analyses if they had either a significant individual effect on the trend or showed a variable-by-time interaction (P <.05). Rate ratios (RRs) based on adjusted analyses were used to test trend differences between groups across time. SAS version 9.4 (SAS Institute, Inc, Cary, NC) was used for all analyses. Study site/geographic region was included in all analyses as a stratification variable, including any possible time-byregion interactions. Data Collection and Analysis for Ursodeoxycholic Acid Treatment and Overall Mortality: Time-to-event outcome and Cox regression model were used to study the impact of treatment on all-cause mortality. Time was measured from the initial PBC diagnosis date (index date) until date of death or last encounter before December 31, Because our earliest index date (2003) also captured patients with pre-existing PBC, we included only patients with index dates of 2004 and later in mortality analyses. Index date covariates included ethnicity (Hispanic, non-hispanic) and age category (40 and [combined], 51 60, 61 70, and >70 y), as well as laboratory results categorized in relation to normal ranges as defined at each site. Supplementary Table 1 describes the categories used for ALP, total bilirubin, albumin, aspartate aminotransferase (AST), alanine aminotransferase (ALT), and the ratio of AST to ALT (ratios 1.1 have been shown to indicate cirrhosis). 9 UDCA treatment (presence/absence) was collected using EHR outpatient prescriptions and pharmacy fill data. To control for differences in demographic and clinical characteristics between patients who did and did not receive UDCA treatment (treatment selection bias), 1 we used a logistic regression model with UDCA as an outcome of interest and baseline variables as covariates to impute an inverse probability of treatment weighting score a weighted propensity-score approach. 10 Treatment selection bias is considered to be controlled if there are no significant differences between patients who did and did not receive treatment after propensity score adjustment. Albumin results were used in propensity score estimation, but excluded from the mortality analyses because of a large proportion of missing data. Multivariable Cox regression with propensity-score adjustment was used to estimate adjusted hazard ratios (ahrs) for all variables retained in the final model. Forest plots were used to illustrate risks/benefits associated with index covariates and UDCA treatment. We performed an ad hoc analysis to test the UDCA benefits across age groups, as well as a sensitivity analysis using a subset of PBC patients (n ¼ 1450) for whom chart abstraction (including UDCA treatment data) was complete. We also studied whether there was an association between ALP levels at diagnosis and index year. Results Trends in Primary Biliary Cholangitis Prevalence: Table 1 and Figures 1 and 2 present yearly unadjusted PBC prevalence rates for the FOLD health system population from 2006 to 2014, overall and by age, gender, and race. Over the study period, PBC prevalence increased significantly from 21.7 to 39.2 per 100,000. Prevalence increased from 33.5 to 57.8 (72% increase) per 100,000 among women, and from 7.2 to 15.4 (114% increase) per 100,000 among men. Similar but smaller increases were observed across age categories and racial groups. No join-point was identified in the prevalence trend, indicating that the increase in prevalence (APC) (Table 2) was steady across time. The overall APC was 5.8% (95% CI, ). When adjusted for region and other covariates, adjusted APC (aapc) differed significantly by age category; aapcs were roughly 7% in the youngest and oldest patients, and 3% to 4% in patients ages 51 to 70,

4 August 2018 Reduced PBC Mortality With Treatment 1345 Table 1. FOLD PBC Prevalence and Incidence by Year: Overall and by Age Category at Index Date, Gender, and Race: Years Population in millions Prevalence Per 100, Age category at index date, y Gender Women Men Race ASINPI Black/African American White Incidence Per 100, Age category at index date, y Gender Women Men Race ASINPI Black/African American White NOTE. Rates per 100,000 persons. indicating that prevalence increased more rapidly among some categories than others. In trend analyses, RRs (Table 2) indicated that the increasing trends in prevalence were consistent within gender and racial groups from 2006 to 2014; prevalence in women remained roughly 4 times as high as in men (RR, 3.8). Likewise, prevalence in African Americans was consistently lower than in white patients (RR, 0.66), and in ASINPI vs white patients (RR, 0.77). However, rate ratios varied across time for age categories, indicating that prevalence changed more in some groups than in others. In 2006, the rate ratio was significantly higher for patients ages 61 to 70 vs those >70 years (RR, 1.3); by 2014, the difference was no longer significant (RR, 1.0). Trends in Primary Biliary Cholangitis Incidence: Figure 1. Overall trends of prevalence and incidence of PBC ( ). Table 1 and Figures 1 and 2 present yearly unadjusted PBC incidence rates, overall and by age, gender, and race. From 2006 to 2014, overall incidence did not vary significantly (from 4.2 to 4.3 per 100,000). Incidence was higher among known risk groups (patients ages 61 70, women, and white patients showed a higher incidence than patients in other age categories, men, and in ASINPI and black patients, respectively), but remained consistent across time for all subgroups.

5 1346 Lu et al Clinical Gastroenterology and Hepatology Vol. 16, No. 8 Figure 2. PBC (A) prevalence and (B) incidence by age, gender, and race. Table 2. Trends in PBC Prevalence and Incidence: Prevalence Incidence RR (95% CI) P value RR (95% CI) P value Variables/risk ratios Gender Women 3.86 ( ) < ( ) <.01 Men Ref Ref Race ASINPI 0.77 ( ) < ( ).54 Black/African American 0.66 ( ) < ( ).01 White Ref Ref in 2006 a Age category at index date, y ( ) < ( ) < ( ) < ( ) < ( ) < ( ) ( ) < ( ) < Ref Ref in 2014 a ( ) < ( ) < ( ) < ( ) Ref Annual percentage change Unadjusted APC 5.8 ( ) < (-1.4 to 1.4).98 Adjusted APC See below 1.6 (-0.2 to 3.5).09 Age-adjusted APC ( ) < ( ) < ( ) ( ) < ( ) <.01 NOTE. Italics indicates the age-adjusted analyses. a Affected by time.

6 August 2018 Reduced PBC Mortality With Treatment 1347 No join-points or trends in incidence were detected over time (unadjusted APC, 0.0% [-1.4 to 1.4]; aapc, 1.6% [-0.2 to 3.5]) (Table 2), indicating that incidence was steady over the study period. RRs for incidence were consistent across time for age, gender, and race categories (Table 2 and Figure 2). The ratio for incidence rates remained roughly 3 times higher for women than for men (RR, 3.2). Ratios were similar between ASINPI and white patients (RR, 1.05), but were significantly lower for African American than for white patients (RR, 0.83). Across the study period, rate ratios for incidence remained steady and significantly higher for patients ages 61 to 70 vs patients >70 years (RR, 1.33), indicating that, despite increasing prevalence among patients >70 years, rates of PBC diagnosis are still highest among patients ages 61 to 70 years. Impact of Baseline Patient Characteristics and Ursodeoxycholic Acid Treatment on All-Cause Mortality A total of 3488 PBC patients had an index date between 2004 and Of them, 2359 patients (70%) had a record indicating any receipt of UDCA and 744 (21%) Table 3. PBC Patient Characteristics and Clinical Condition at Baseline ( ) Variable UDCA never (N ¼ 1129) UDCA ever (N ¼ 2359) Before propensity score weighting, P value After propensity score weighting, P value Region Midwest 249 (22%) 521 (22%) Northeast 103 (9%) 205 (9%) South 184 (16%) 186 (8%) West 593 (53%) 1447 (61%) Gender Women 843 (75%) 2007 (85%) < Men 286 (25%) 352 (15%) Hispanic ethnicity No 211 (19%) 510 (22%) < Yes 622 (55%) 1498 (64%) Unknown 296 (26%) 351 (15%) Age at index, y (8%) 190 (8%) < (13%) 363 (15%) (26%) 743 (31%) (28%) 670 (28%) (25%) 393 (17%) Race ASINPI 82 (7%) 170 (7%) < Black/African American 134 (12%) 152 (6%) White 633 (56%) 1594 (68%) Unknown 280 (25%) 443 (19%) ALP level Normal 230 (20%) 281 (12%) < *ULN 370 (33%) 616 (26%) 2 3*ULN 96 (9%) 266 (11%) 3*ULN 107 (9%) 361 (15%) Unknown 326 (29%) 835 (35%) Albumin level Normal 453 (40%) 909 (39%) <LLN 228 (20%) 274 (12%) Unknown 448 (40%) 1176 (50%) Total bilirubin level 0 < (49%) 1126 (48%) < (9%) 189 (8%) (4%) 74 (3%) (11%) 184 (8%) Unknown 303 (27%) 786 (33%) AST/ALT ratio 1.1 No 483 (43%) 1039 (44%) < Yes 383 (34%) 514 (22%) Unknown 263 (23%) 806 (34%) NOTE. Asterisk (*) is used as a multiplication symbol (1-2 times the upper limit of normal). LLN, lower limit of normal; ULN, upper limit of normal.

7 1348 Lu et al Clinical Gastroenterology and Hepatology Vol. 16, No. 8 had died. Table 3 presents patient characteristics by UDCA treatment status (ever-treated vs never-treated). After propensity score adjustment, there were no significant differences in treatment status by patient characteristics. Multivariate Cox regression (Figure 3) showed that age at index date, gender, ALP level, bilirubin level, AST/ ALT ratio, and UDCA treatment status were associated significantly with overall mortality. Mortality was higher in men than in women (ahr, 1.48 [95% CI, ]). Likelihood of death increased with ALP level increase at index date; patients with ALP levels 3 times the upper limit of normal were more than twice as likely to die as patients with a normal ALP level. Likewise, patients with increased bilirubin level (categories >1.0 mg/ml) were 2 to 3 times more likely to die than patients with lower bilirubin levels (<1.0 mg/ml). Patients with AST/ALT ratios 1.1 were at 2.6 times the risk of death than patients with lower ratios (ahr, 2.57 [ ]). Race was not associated significantly with mortality. Treatment with UDCA was associated with significantly reduced mortality (ahr, 0.57; 95% CI, ). These benefits were observed in all age groups (ahrs range, ) (Supplementary Methods section and Supplementary Table 1). A subgroup analysis (n ¼ 1450) based on patients with completed chart abstraction showed that more than 79% of patients had been treated with UDCA and showed an even stronger effect of UDCA status on mortality; after adjustment for other covariates, treatment was associated with a 52% reduction in mortality (ahr, 0.48; 95% CI, ). We also performed an ad hoc analysis in which we observed that, over the study period, average ALP levels were lower as index years (date of diagnosis) progressed, suggesting diagnoses earlier in the disease process. Discussion From 2006 to 2014, we observed a significant increase in the prevalence of PBC, without a concomitant increase in incidence. This phenomenon suggests that patients are living longer with the disease and/or being diagnosed earlier in the disease process. We did observe that prevalence increased more rapidly among patients 50 and >70 years old than among patients ages 51 to 70 years. In an ad hoc analysis, we also found a trend toward patients being diagnosed with less increased ALP levels across the study period (data not shown). Altogether, our findings suggest improvements in both early diagnosis and reduced mortality among PBC patients under routine clinical care in the United States. PBC prevalence and incidence among white and ASINPI patients in our US cohort were higher than those reported in Europe and Asia, respectively, among studies using analytical approaches similar to our own. 2,3 Nevertheless, our findings are consistent with those of other US-based studies. For example, we observed prevalence rates in white patients (29.0 per 100,000 in 2006, and 46.0 per 100,000 in 2014) that were comparable with those reported in a primarily white population from Olmsted County, Minnesota (40.2 per 100,000). 11 Given that our sample was drawn from a geographically diverse group of nontertiary care health systems, Figure 3. Multivariable Cox regression: ahrs for all-cause mortality. UDCA treatment status is propensity score adjusted (see Table 3). Bilirubin units are provided in IU/ ml. Normal was defined by the manufacturer of the assay used at each site. Age category is defined at index date. LCL, lower confidence limit; UCL, upper confidence limit; ULN, upper limit of normal. *, 1-2 times the upper limit of normal.

8 August 2018 Reduced PBC Mortality With Treatment 1349 our incidence and prevalence estimations may be generalizable to the US population. After age category, gender was the strongest patient characteristic related to PBC prevalence and incidence (Figures 2 and 3). The ratios of women to men in both prevalence (3.9:1) and incidence (3.2:1) were stable across time; these findings also were consistent with recent reports from Europe. 3 A similar incidence ratio (4:1) was observed in The Netherlands 12 from 2000 to However, a recent study from South Korea reported a much higher incidence ratio between women and men (7.3:1); in that study, the RRs also remained consistent for women vs men across time. 2 We report trends in PBC epidemiology among African Americans under routine clinical care. The prevalence and incidence of PBC among African American patients was lower than in white patients in our cohort, but was not insignificant. It is not clear if these findings reflect a difference in susceptibility or a lack of clinician awareness regarding the occurrence of PBC in African Americans. 13 Goldberg et al 14 recently observed that another autoimmune biliary disease primary sclerosing cholangitis likewise may be underdiagnosed among African Americans. We detected region-by-time interactions in PBC incidence. However, interpretation of these findings is complicated by the dynamic nature of the site variable. Health systems may undergo restructuring and merging; accepted health insurance programs (private as well as Medicaid) also may change over time, affecting the profile of patients seen within each system. For these reasons, we included geographic region as a stratification variable rather than considering it a risk factor. We believe there is a benefit ofudcatreatmenton all-cause mortality in PBC patients under routine care. After propensity score adjustment to control for treatment selection bias, and adjustment for region and other index covariates, we found that gender, ALP level, total bilirubin level, AST/ALT ratio, and UDCA treatment were independent risk factors for all-cause mortality. Our analysis showed a 43% reduction in mortality among UDCA-treated patients vs those who were untreated (ahr, 0.57; 95% CI, ). This UDCA effect was confirmed by a sensitivity analysis based on a subgroup of PBC patients with detailed data drawn from complete chart abstraction; after adjustment for other covariates, treatment was associated with a 52% reduction in overall mortality (ahr, 0.48; 95% CI, ). Although our earlier research found that African American PBC patients were less likely to receive UDCA treatment, 1 race was not a significant predictor for overall mortality in the current analysis. Our analysis also showed that men with PBC were at higher risk of death than women (HR, 1.3 [95% CI, ]). This finding is consistent with results from Europe 3 and Canada. 15 However, the effect we observed was weaker than that found in these other studies. One explanation is that our analysis has adjusted for clinical factors, such as age category at index date. In our earlier report, we observed that men were less likely to receive UDCA, and less likely to respond to UDCA when treated. Increased ALP and bilirubin levels have long been recognized as markers of PBC severity 9 ; consistent with this observation, we found that these factors were associated with an increased risk of death in our cohort. Likewise, our finding that age at diagnosis was associated with higher mortality also is consistent with previous findings. 16 A ratio of AST/ALT 1.1 has been shown to be an indicator of cirrhosis in PBC patients. 9 We believe our study shows that this increased ratio is also a risk factor for all-cause mortality. Future analyses will examine whether a more comprehensive investigation of laboratory results, alone or in combination (eg, GLOBE scores 17 ), as well as detailed UDCA treatment (eg, dose, duration), will have prognostic utility. An unavoidable limitation of any health system based cohort study is that persons without health coverage are not represented. However, the FOLD health systems include privately insured, Medicare, and Medicaid patients, thereby representing a broad cross-section of the socioeconomic distributions of their geographic areas. Given that our sample was drawn from a geographically diverse group of nontertiary care health systems, our estimates may be generalizable to the US population. Another limitation of our study was that there was a large proportion of missing data at the index date. To account for this, missing data were included as a category (unknown) in both propensity score matching and multivariable analyses of UDCA treatment status for all patients. Although missing data may complicate propensity score adjustment for treatment selection bias, the proportions of unknowns were comparable between treated and untreated patients, and we believe it was unlikely to have impacted our results significantly. In a sensitivity analysis, we found that patients of unknown race were younger and less likely to be treated with UDCA, but we did not observe any differences in clinical markers of PBC severity such as ALP and bilirubin. It is possible that unknown laboratory test results, especially for ALP, could be owing to low awareness of PBC among primary care physicians and/or a lack of patient access to specialty providers who treat PBC patients. Interestingly, a recent cohort study of individual PBC patients from 15 North American and European sites found that roughly a quarter of patients were missing ALP and bilirubin results. 18 We expect that ongoing chart abstraction will greatly reduce missing data in future FOLD analyses. Our preliminary data set contained only ever/never use of UDCA; future analyses will include more comprehensive treatment data, including timing, dosage, and duration of use, as well as investigate any possible interactions between UDCA and other variables. We also note that, although the rate of false-positive PBC cases identified by our automated algorithm had little impact

9 1350 Lu et al Clinical Gastroenterology and Hepatology Vol. 16, No. 8 on our prevalence estimates, 1 it is likely that any such cases fell into the never-treated category, potentially inflating the proportion of patients we observed who did not receive UDCA. Our subgroup analyses of patients with completed chart abstraction found that 79% of patients had ever received UDCA treatment. This is comparable with the 85% treatment rate observed in the Global PBC study group, which is dominated by referral liver centers. 18 The slightly lower treatment rates we observed may reflect our consortium s routine clinical care setting; only 3 sites (serving 16% of patients) include liver specialty centers. 1 In summary, the FOLD Consortium characterizes trends in PBC prevalence and incidence as well as the impact of UDCA treatment on mortality across a geographically and racially diverse sample of US patients. Our findings suggest improvements in reduced mortality, especially in UDCA-treated patients, as well as earlier diagnosis among PBC patients under routine clinical care in the United States. Supplementary Material Note: To access the supplementary material accompanying this article, visit the online version of Clinical Gastroenterology and Hepatology at and at References 1. Lu M, Li J, Haller IV, et al. Factors associated with prevalence and treatment of primary biliary cholangitis in United States health systems. Clin Gastroenterol Hepatol 2018;16: Kim KA, Ki M, Choi HY, et al. Population-based epidemiology of primary biliary cirrhosis in South Korea. Aliment Pharmacol Ther 2016;43: Lleo A, Jepsen P, Morenghi E, et al. Evolving trends in female to male incidence and male mortality of primary biliary cholangitis. Sci Rep 2016;6: Saffioti F, Gurusamy KS, Eusebi LH, et al. Pharmacological interventions for primary biliary cholangitis: an attempted network meta-analysis. Cochrane Database Syst Rev 2017;3: Cd Lindor KD, Therneau TM, Jorgensen RA, et al. Effects of ursodeoxycholic acid on survival in patients with primary biliary cirrhosis. Gastroenterology 1996;110: Jarow JP, LaVange L, Woodcock J. Multidimensional evidence generation and FDA regulatory decision making: defining and using real-world data. JAMA 2017;318: Breiman L, Friedman J, Stone Charles J. Classification and regression trees. 1st ed. New York: Chapman and Hall, Kim HJ, Fay MP, Feuer EJ, et al. Permutation tests for joinpoint regression with applications to cancer rates. Stat Med 2000; 19: Nyblom H, Bjornsson E, Simren M, et al. The AST/ALT ratio as an indicator of cirrhosis in patients with PBC. Liver Int 2006; 26: Steventon A, Grieve R, Sekhon J. A comparison of alternative strategies for choosing control populations in observational studies. Health Serv Outcomes Res Method 2015;15: Kim WR, Lindor KD, Locke GR 3rd, et al. Epidemiology and natural history of primary biliary cirrhosis in a US community. Gastroenterology 2000;119: Boonstra K, Kunst AE, Stadhouders PH, et al. Rising incidence and prevalence of primary biliary cirrhosis: a large populationbased study. Liver Int 2014;34:e31 e Toy E, Balasubramanian S, Selmi C, et al. The prevalence, incidence and natural history of primary sclerosing cholangitis in an ethnically diverse population. BMC Gastroenterol 2011; 11: Goldberg D, Levy C, Yimam K, et al. Primary sclerosing cholangitis in black patients is not rare in a multi-center North American Consortium. Clin Gastroenterol Hepatol 2017 Nov 2. Epub ahead of print. 15. Myers RP, Shaheen AA, Fong A, et al. Epidemiology and natural history of primary biliary cirrhosis in a Canadian health region: a population-based study. Hepatology 2009;50: Corpechot C, Chazouilleres O, Poupon R. Early primary biliary cirrhosis: biochemical response to treatment and prediction of long-term outcome. J Hepatol 2011;55: Lammers WJ, Hirschfield GM, Corpechot C, et al. Development and validation of a scoring system to predict outcomes of patients with primary biliary cirrhosis receiving ursodeoxycholic acid therapy. Gastroenterology 2015;149: e Lammers WJ, van Buuren HR, Hirschfield GM, et al. Levels of alkaline phosphatase and bilirubin are surrogate end points of outcomes of patients with primary biliary cirrhosis: an international follow-up study. Gastroenterology 2014;147: e5. Reprint requests Address requests for reprints to: Mei Lu, PhD, Department of Public Health Sciences, Henry Ford Health System, 3E One Ford Place, Detroit, Michigan mlu1@hfhs.org; fax: (313) Conflicts of interest These authors disclose the following: Stuart C. Gordon has received grant/ research support from AbbVie Pharmaceuticals, Bristol-Myers Squibb, Conatus, CymaBay, Exalenz BioScience, Gilead Pharmaceuticals, Intercept Pharmaceuticals, and Merck, and serves as an ad hoc consultant/advisor for Abbvie, Bristol-Myers Squibb, CVS Caremark, Dova Pharmaceuticals, Gilead, Intercept, and Merck; Mei Lu, Jia Li, Lora Rupp, Sheri Trudeau, Talan Zhang, Yueren Zhou, Yihe G. Daida, Mark A. Schmidt, and Joseph A. Boscarino have received grant/research support from Gilead Pharmaceuticals; Carla V. Rodriguez has received grant/research support from Merck and owns stock in Gilead (<$5000); Heather Anderson has received grant/research support from Intercept Pharmaceuticals; Jeffrey J. VanWormer has received grant/research support from Retrophin; Christopher L. Bowlus has received grant/research support from AbbVie Pharmaceuticals, Bristol-Myers-Squibb, Gilead Biosciences, Intercept Pharmaceuticals, Merck, Shire Pharmaceuticals, and Takeda Pharmaceuticals, and has served as an advisor for Bristol-Myers- Squibb, Gilead Biosciences, Intercept Pharmaceuticals, and Takeda; Keith Lindor is a consultant/advisor for HighTide Biopharma, has served as an ad hoc advisor for Takeda, Shire, and Intercept Pharmaceuticals, and sits on a Data Safety Monitoring Board for Takeda; and Robert J. Romanelli has received grant/research support from Pfizer, Inc, and Janssen Scientific Affairs. The remaining authors disclose no conflicts. Funding Supported by Intercept Pharmaceuticals, Inc.

10 August 2018 Reduced PBC Mortality With Treatment 1350.e1 Supplementary Methods We performed a brief analysis and found an interaction between patient age and treatment with UDCA in the effect on mortality (P ¼.02). UDCA treatment was associated with reduced mortality in all age groups; the effect was strongest among the youngest patients. Supplementary Table 1. Hazard Ratios for Mortality Among UDCA-Treated vs Untreated Patients Age category, y HR 95% Wald CI > NOTE. P value for interaction ¼.02.