Loss of ambulation (LoA) is a critical milestone in the progression of Duchenne muscular dystrophy (DMD) that usually occurs by the early teens, with significant variation across individuals. Accurate individualized predictions of time to LoA are valuable for DMD drug development (e.g., enrichment of trial populations and benchmarking of long-term treatment outcomes), as well as for counseling of patients and caregivers (e.g., advance ordering of a power wheelchair, preparing home modifications for accessibility). A composite prognostic score for LoA was previously developed based data from 608 boys and 1,240 years of follow-up across multiple DMD natural history and placebo-arm data sources. Timed function tests, steroid type, height and weight were identified as the most important predictors of LoA. We independently validated this score using data from 226 patients with 819 years of follow-up in the CINRG Duchenne Natural History Study, which was not used for score development. The previously-defined prognostic groups showed excellent risk stratification in the validation data, with median times to LoA of 1.8, 3.8, 7.8 and >9.0 years across groups, compared with medians of 1.1, 2.1, 4.1 and >9.0 years in the training data. Risk groups were well-differentiated despite the use of different LoA definitions in the training data (inability to walk 10 meters or complete the 6-minute walk test) and validation data (full-time wheelchair dependence), but shifted to later median ages in the validation data. Simplified prognostic scores were also explored using classification and regression trees to identify functional thresholds that best predicted time to LoA. Individualized prediction of time to LoA is possible in DMD, and is most accurate when based on a composite profile of motor function. Calibration of prognostic scores to specific definitions of LoA is important for quantitative accuracy, and for applications in clinical management and drug development.