Objectives: To estimate incidence, identify consequences and potential risk factors for harm in people with dementia who got lost in one UK policing region. Methods: In a retrospective observational study, data were extracted from missing-person records over a four-year period in one UK policing region (population of 2.1 million). Results: Two hundred and eighty-one incidents of getting lost were identified. Incidence of getting lost was estimated at 0.5% of the regional dementia population. Fifty-nine percent of reports came from domestic settings, 29% from care homes/hospitals, and 12% on excursions from home. Five percent (n = 15) sustained significant harm, including two deaths. Average age was 78 years (SD 8.3). Harm was associated with older age (mean difference 6.16 years, CI 1.86 to 10.46, p = 0.005, t = 2.82), length of time missing (Mdn time 2.48 hours; IQR 0.97 to 9.45, p = 0.02), and season (9% winter, 2% summer, p = 0.006). The length of time missing increased with delays in reporting to police (r = 0.15, p = 0.018), getting lost at night (Mdn time 1.70 hours, IQR 0.52-3.32, p = 0.028), driving themselves (Mdn time 2.45 hours, IQR 0.42-2.00, p = 0.001), and using public transport (Mdn 1.78 hours, IQR 1.07-3.92, p = 0.001). Conclusion: Incidence in this study suggests getting lost is a low-frequency event for people with dementia but for a small minority, the risks are considerable. Exploratory analyses suggest individual and environmental factors increase the risk of harm. Suitable methods need to be developed to replicate these findings in larger prospective samples. A focus on the predictors of harm may aid development of assessment protocols to ensure intervention is proportionate.