**Background:** Missing data in health surveys can threaten the validity of research conclusions, especially in large cohort studies aiming to include diverse populations. The All of Us Research Program collects extensive survey data to advance precision medicine, but patterns of missingness in its baseline surveys had not been systematically described. Understanding these patterns is critical for researchers using the data.
**Methods:** The study analyzed survey responses from 334,183 participants who completed at least one of three baseline surveys (The Basics, Overall Health, Lifestyle) between May 31, 2017, and September 30, 2020. Missingness was defined as item nonresponse (participant saw a question but did not answer). Questions with branching logic were excluded if not applicable. The primary outcome was the missing percentage (ratio of skipped to eligible questions). Explanatory variables included race/ethnicity, education, income, sexual/gender minority status, geography (urban vs. rural), age, health literacy score (brief health literacy scale, range 3–15), and enrollment date. Missing percentages were compared using Wilcoxon rank-sum/Kruskal-Wallis tests, and associations with continuous variables were assessed via Spearman correlation. Negative binomial regression (with five-knot natural cubic splines for age and enrollment time) was used to model the number of missed questions, adjusting for all covariates. Multiple imputation (mice package, 6 imputations) addressed missing health literacy scores (~6% missing). Sensitivity analyses included complete-case analysis, alternative imputation methods, and counting “Prefer not to answer”/“Don’t know” as missing.
**Key Results:** Among 334,183 participants, 97.0% completed all three baseline surveys, and only 0.2% skipped all questions in at least one survey. The median skip rate was 5.0% (IQR 2.5%–7.9%); 74.9% skipped fewer than 10% of questions, and 0.2% skipped more than half. Historically underrepresented groups had higher missingness: Black/African American participants had an incidence rate ratio (IRR) of 1.26 (95% CI: 1.25–1.27) compared to White participants; Hispanic/Latino IRR 1.15 (1.14–1.16); other/multiple races IRR 1.22 (1.21–1.23). Lower education was associated with higher missingness: less than high school IRR 1.14 (1.13–1.15); high school IRR 1.11 (1.10–1.12) vs. college degree. Skipping specific demographic questions was strongly associated with overall missingness: skipping income IRR 1.39 (1.38–1.40); skipping race/ethnicity IRR 1.69 (1.66–1.72); skipping education IRR 1.92 (1.89–1.95); skipping sexual/gender questions IRR 2.19 (2.09–2.30). Rural geography had slightly lower missingness (IRR 0.93, 0.92–0.94). Missingness was relatively stable over time, age, and health literacy score, with minor deviations around national launch (May 2018) and COVID-19 onset (March 2020). Sensitivity analyses confirmed primary results.
**Clinical Implications:** The low overall missingness in All of Us baseline surveys is reassuring for researchers, but the differential missingness by race/ethnicity, education, and other demographics poses a risk of bias if not addressed. Participants who skip key demographic questions are much more likely to skip other questions, suggesting that these items can serve as “leading indicators” of missingness. Researchers should avoid complete-case analyses without assessing missingness patterns, and consider using multiple imputation or weighting to mitigate bias. The findings highlight the need for careful survey design and statistical adjustment in large, diverse cohorts to ensure that the inclusion of underrepresented groups does not lead to biased conclusions due to differential nonresponse.