For diagnostic research, this implies that multiple clinically relevant diagnoses need to be considered, particularly as there is a large overlap of symptoms in these diseases (including, for example, shortness of breath, chest pain, or reduced exercise tolerance). Current diagnostic studies typically focus on predicting the presence or absence of one single target condition, even though multiple diagnoses are often considered in daily practice. There are, however, statistical techniques available that better follow the way physicians diagnose disease in daily patient care, working with a broad differential diagnosis of multiple diseases instead of only one disease. Epidemiologic studies thus need to better follow clinical medicine and try to provide evidence on how to simultaneously predict two or more conditions, instead of only one. Hereto, so-called polytomous regression models can be used.14,15 In technical terms, such polytomous modeling fits multiple submodels within the same maximum likelihood estimation, comparing each differential diagnosis with a chosen reference diagnosis. Per disease included in the differential diagnosis, different regression coefficients (or ORs) are estimated per diagnostic test, allowing that particular test to have varying diagnostic accuracy for different diagnoses. As an example, in a 72-year-old man presenting with shortness of breath and coughing, such a model can directly provide the probability of both COPD and heart failure, given his unique patient profile of a combination of a slightly elevated N-terminal prohormone brain natriuretic peptide level, a history of smoking, and some signs of fluid overload. Next, this can guide referral for additional tests and treatment options. Unfortunately, such models that allow to simultaneously predict the presence (or absence) of multiple diseases are still not very common in the diagnostic literature. Possible reasons may be their increased complexity, lack of experience with the method, lack of adequate data on all relevant outcome categories, or the force of habit, with lack of data perhaps being one of the more difficult barriers to take down. To construct robust polytomous models, obviously, enough data need to be available on both the diagnostic tests and the presence of all relevant diseases. Moreover, one should ensure sufficient number of patients in all outcome categories to facilitate statistical modeling. Combining individual patient data (IPD) from multiple studies to increase sample size and robustness of findings is a very attractive solution for this purpose. Novel methodology on how to perform these so-called IPD meta-analyses is increasingly being developed, and its use is helpful not only for the construction of advanced diagnostic models but also to evaluate the effect of comorbidity on the performance of diagnostic or prognostic tests or whether the (side) effect of a treatment differs according to relevant covariates, such as comorbidity or increasing age.16 In subsequent sections we give more detail on some important aspects of how to perform IPD meta-analyses and give some examples on how this may help improve patient care for patients with multimorbidity.