Introduction Databases such as the National Cancer Database (NCDB) are important for analyzing care provisions and clinical outcomes. However, missing data limits the scope and skews the accuracy of analyses. It may be correlated with the quality of care at the institution. This study aims to assess the prevalence of and characteristics of missing data in the NCDB for gallbladder, biliary, and hepatic cancers.

Methods This retrospective cohort study reviewed 12 key variables within the NCDB for missing values for patients diagnosed with a hepatic, biliary, or gallbladder cancers from 2010 to 2019. Key variables included demographic (age, race, ethnicity, insurance, education, income, comorbidities), tumor (grade, TNM stages), and treatment characteristics (surgery, chemotherapy, radiation therapy). The percentage of patients with complete and missing number of key variables was analyzed. Characteristics of the missing data including the distribution of missing data by cancer stage, percentage of patients with missing data by year, and the percentage of patients missing per number of studied variables was calculated.

Results A total of 308,354 patients (mean SD age, 66.3 11.5 years; 197,876 64.2% male) in the NCDB were identified; 13.1% were missing at least 1 key variable. Missing data was similarly prevalent amongst the cancers (12.2% in gallbladder, 13.25% in liver, and 13.26% in biliary cancers). Demographic data was most likely to be missing in biliary cancers (13.2%) and tumor data was most likely to be missing in liver cancers (0.12%), though very little tumor data was missing. Of those missing data, most patients were missing seven key variables. Gallbladder and biliary cancers are most likely to be missing stage 4 data, while liver cancers are more likely to be missing stage 1 cancer. The percentage of patients with missing data has increased from 2010-2019.

Conclusion Missing data of select variables in the NCDB prevalent and the percentage of missing key variables is similar across hepatic, biliary, and gallbladder cancers. Understanding that the patients missing data are not random impacts the generalizability of NCDB analysis.