Predicting risk of chronic liver disease in Chinese adults: External validation of the CLivD score

Development and validation of a model to predict incident chronic liver disease in the general population: The CLivD scoreJournal of HepatologyVol. 77Issue 2PreviewCurrent screening strategies for chronic liver disease focus on detection of subclinical advanced liver fibrosis but cannot identify those at high future risk of severe liver disease. Our aim was to develop and validate a risk prediction model for incident chronic liver disease in the general population based on widely available factors. Full-Text PDF Open Access Chronic liver disease (CLD), including cirrhosis and liver cancer, accounted for 9.5 million deaths globally in 2017, one-third of which occurred in China.[1]GBD 2017 Causes of Death CollaboratorsGlobal, regional, and national age-sex-specific mortality for 282 causes of death in 195 countries and territories, 1980-2017: a systematic analysis for the Global Burden of Disease Study 2017.Lancet. 2018; 392: 1736-1788https://doi.org/10.1016/S0140-6736(18)32203-7Abstract Full Text Full Text PDF PubMed Scopus (4669) Google Scholar Non-alcoholic fatty liver disease (NAFLD) alongside its metabolic risk factors has become a major risk factor for CLD.[2]Xiao J. Wang F. Wong N.K. et al.Global liver disease burdens and research trends: analysis from a Chinese perspective.J Hepatol. 2019; 71: 212-221https://doi.org/10.1016/j.jhep.2019.03.004Abstract Full Text Full Text PDF PubMed Scopus (344) Google Scholar Based on easily accessible risk factors, the CLivD score, a simple model to predict future risk of CLD, has been developed and validated in European populations.[3]Åberg F. Luukkonen P.K. But A. et al.Development and validation of a model to predict incident chronic liver disease in the general population: the CLivD score.J Hepatol. 2022; 77: 302-311https://doi.org/10.1016/j.jhep.2022.02.021Abstract Full Text Full Text PDF PubMed Scopus (17) Google Scholar This model can be used to identify individuals at high risk who could benefit from lifestyle interventions and should be referred for further liver assessment. However, the associations of lifestyle and metabolic risk factors with CLD risk differ in Chinese adults.[4]Pang Y. Kartsonaki C. Turnbull I. et al.Diabetes, plasma glucose, and incidence of fatty liver, cirrhosis, and liver cancer: a prospective study of 0.5 million people.Hepatology. 2018; 68: 1308-1318https://doi.org/10.1002/hep.30083Crossref PubMed Scopus (115) Google Scholar,[5]Pang Y. Kartsonaki C. Lv J. et al.Observational and genetic associations of body mass index and hepatobiliary diseases in a relatively lean Chinese population.JAMA Netw Open. 2020; 3e2018721https://doi.org/10.1001/jamanetworkopen.2020.18721Crossref Scopus (11) Google Scholar It is unclear whether the CLivD score could be used in Chinese adults after model recalibration. Therefore, we performed an external validation of the CLivD score for Europeans in the China Kadoorie Biobank (CKB). The CKB recruited 512,715 participants aged 30-79 years from 10 areas (5 urban, 5 rural) in China during 2004-2008.[6]Chen Z. Chen J. Collins R. et al.China Kadoorie Biobank of 0.5 million people: survey methods, baseline characteristics and long-term follow-up.Int J Epidemiol. 2011; 40: 1652-1666Crossref PubMed Scopus (658) Google Scholar All participants completed an interviewer-administered laptop-based questionnaire and physical measurements were recorded by trained technicians. Seventeen biochemistry traits were measured in a subsample of 17,681 participants.[7]Wang X. Wu Z. Lv J. et al.Life-course adiposity and severe liver disease: a Mendelian randomization analysis.Obesity. 2023; 31: 3077-3085https://doi.org/10.1002/oby.23913Crossref PubMed Scopus (0) Google Scholar Follow-up for ICD-10-coded incident events continued to January 1, 2018, through linkage with morbidity and mortality registries and a nationwide health insurance system. CLD was defined as advanced liver disease or liver-related mortality, including cirrhosis, NAFLD, and liver fibrosis (ICD-10 code: K70, K72, and K74 alongside other complications of CLD).[3]Åberg F. Luukkonen P.K. But A. et al.Development and validation of a model to predict incident chronic liver disease in the general population: the CLivD score.J Hepatol. 2022; 77: 302-311https://doi.org/10.1016/j.jhep.2022.02.021Abstract Full Text Full Text PDF PubMed Scopus (17) Google Scholar Discrimination was assessed using Harrell's C index. Calibration was assessed graphically by comparing the predicted risks and the observed risk. We recalibrated the original CLivD score model by re-estimating the predictor coefficients in CKB (i.e. model refit).[8]Van Calster B. McLernon D.J. van Smeden M. Wynants L. Steyerberg E.W. Topic Group 'Evaluating diagnostic tests and prediction models' of the STRATOS initiative. Calibration: the Achilles heel of predictive analytics.BMC Med. 2019; 17: 230https://doi.org/10.1186/s12916-019-1466-7Crossref PubMed Scopus (626) Google Scholar Of the 504,009 participants included (excluding participants with a prior history of cancer, cirrhosis or hepatitis), the mean age was 52 years (SD 10.7 years), and 59.2% were women. During 11 years of follow-up, there were 4,091 incident cases of CLD. The Harrell's C index of the CLivD score was 0.65 (0.64-0.66) for modelnon-lab and 0.68 (0.62-0.73) for modellab (Fig. 1). The Harrell's C index for modelnon-lab in CKB was similar to those reported by the original CLivD score paper, but the Harrell's C index for modellab in CKB was lower. For modelnon-lab, the Harrell's C index was slightly lower among men and participants who tested positive for HBsAg and those with diabetes. For modellab, the Harrell's C index was slightly higher among women and those from rural areas, weekly drinkers, and those with a fatty liver index ≥60 (Fig. 1). Both models under-estimated risk at lower levels of observed risk but over-estimated risk among participants among the top decile of observed risk. After recalibration, both models still underestimated the risk (Fig. 1). Modelnon-lab had good overall discrimination in this Chinese population. Three prospective studies have externally validated the CLivD score. The first two studies were European cohorts and reported a C-statistic of 0.65 (CCHS) and 0.74 (Whitehall II) for modelnon-lab, separately.[6]Chen Z. Chen J. Collins R. et al.China Kadoorie Biobank of 0.5 million people: survey methods, baseline characteristics and long-term follow-up.Int J Epidemiol. 2011; 40: 1652-1666Crossref PubMed Scopus (658) Google Scholar The third study was NHANES, involving diverse race and ethnicities, and showed a C-statistic of 0.637 for modelnon-lab[9]Song J. Jiang Z.G. A good step toward low-cost prognostication of liver-related outcome awaits more validation.J Hepatol. 2022; 77: 887-889https://doi.org/10.1016/j.jhep.2022.04.008Abstract Full Text Full Text PDF PubMed Scopus (1) Google Scholar. The current study reported a C-statistic of 0.649 for modelnon-lab, which was generally consistent with the aforementioned studies. In contrast, modellab performed worse in the Chinese population than previous studies. The C-statistic for modellab was 0.78 in CCHS and 0.733 in NHANES, respectively, and was higher than the C-statistic of 0.68 in CKB. Possible reasons for the worse performance of modellab included the much smaller number of participants with biochemistry data in CKB, their inclusion criteria and the weak associations between predictors and incident CLD. The CLivD score under-estimated the risk of CLD in CKB across the range of observed risk even after recalibration by refitting the models. Underestimation was a particular issue in HBsAg-positive participants, which is anticipated since progression of liver disease is driven more by viral factors than the factors included in the CLivD score. In contrast, calibration was much better in the HBsAg-negative individuals. The unsatisfying performance of model calibration may reflect the fact that different sets of predictors are needed for Chinese individuals to develop risk prediction models with good calibration as well as good discrimination. This work was supported by the National Key R&D Program of China (2023YFC3606302) and National Natural Science Foundation of China (82192904, 82192901, 82192900). The CKB baseline survey and the first re-survey were supported by a grant from the Kadoorie Charitable Foundation in Hong Kong. The long-term follow-up is supported by grants from the UK Wellcome Trust (212946/Z/18/Z, 202922/Z/16/Z, 104085/Z/14/Z, 088158/Z/09/Z), grants from the National Key R&D Program of China (2016YFC0900500), National Natural Science Foundation of China (81390540, 91846303, 81941018), and Chinese Ministry of Science and Technology (2011BAI09B01). The authors have no relevant financial or non-financial interests to disclose. Please refer to the accompanying ICMJE disclosure forms for further details. Conceptualization, Y.P. and C.K.; methodology, Y.P. and C.K.; software, Y.P., C.K., and F.A.; validation, Y.P., C.K., and F.A.; formal analysis, Y.P. and C.K.; investigation, Y.P.; resources, Y.P., C.K., and F.A.; data curation, Y.P. and C.K.; writing—original draft preparation, Y.P. and C.K.; writing—review and editing, Y.P., C.K., F.A., L.L., and Z.C.; visualization, Y.P. and C.K.; supervision, L.L. and Z.C.; project administration, Y.P., C.K., L.L., and Z.C.; funding acquisition, Y.P., C.K., L.L., and Z.C. All authors have read and agreed to the published version of the manuscript. The study was approved by the Institutional Review Board of Peking University Health Science Center (ID of the approval: IRB00001052-20040). Informed consent was obtained from all participants. The following are the supplementary data to this article: Download .pdf (.59 MB) Help with pdf files Multimedia component 1