Randomization is the ideal way to evaluate alternative treatment approaches but is often not feasible for practical and ethical reasons. International differences in the approach to treating nursing home residents with dementia provide an important opportunity to investigate the consequences of palliative versus curative approaches on comfort and survival. However, our prior work has identified three important methodological issues with such cross-cultural observational studies. First, identification of relevant decision-making may not be complete. For this, we propose a two-way approach with physicians' identification of status change events, and families' regular reporting on decision-making. Second, without randomization, control for confounding by differing patient characteristics and reasons for assigning different treatments (confounding by indication) are a major issue. We therefore propose collecting extensive data on patient, family, physician, and facility factors to adjust for confounders for the association between treatment approach and outcomes (comfort and survival). Nevertheless, in spite of this, residual confounding might persist. Third, instruments should be valid across nations. We recommend using a recently developed instrument, the Minimum Data Set-Palliative Care. Reliability and validity of translations are being tested. We will illustrate these three issues with results of our study on effects of pneumonia treatment in US and Dutch nursing home residents. In spite of potential and experienced methodological problems, the proposed design, careful adjustment for confounders, and awareness to potential cross-cultural differences provide us with the best opportunity to investigate treatment effects in frail patients with advanced dementia.