Development and validation of a risk prediction model for visual impairment in older adults.

Single centre in one Chinese province, so the results may not transfer to other populations, health systems or ethnic groups. The population was drawn from ophthalmology outpatients and a physical examination centre, skewing toward people already seeking eye care, so the findings cannot be generalised to the community. Case-control design with data collected at one time point cannot establish the sequence of exposure and outcome, so it cannot show that any factor caused visual impairment. Convenience sampling was used, not random selection of the whole eligible population. Internal validation only, with no external dataset, no calibration assessment and no reported sensitivity or specificity at a usable threshold, so the score's real-world usefulness is unknown. Several variables were self-reported (diabetes history, activity, sleep), and some predictors were dropped for being hard to measure, so the model may miss relevant risk. The model was built on a hospital sample where the outcome was defined by eye examination, which inflates discrimination compared with screening people in the community. Authors state the findings cannot be generalised to community, province or national populations.

The authors declare no conflicts of interest. Data collection costs such as materials and transport were funded by the National Natural Science Foundation of China. No commercial sponsor designed or wrote the study.

Do not treat the AUC of 0.81 as proof this score works for screening older adults in your clinic. That number came from internal validation in a hospital population already attending eye services, with no external testing and no calibration, so it may look strong only because cases and controls were drawn from settings that differ. The associations with blood pressure, activity and education also cannot be read as causes, because this was a case-control study that measured exposure and outcome at the same time.