Correlate diseases with their symptoms and signs
Answers
Answer:
Hmmm..
Explanation:
Many temporal correlations between specific disease states are well recognized. For example, hypertension may increase the risk of heart disease.1 Traditionally, such temporal correlations have been identified through epidemiological studies focusing on specific pairs of diseases. With the availability of large-scale EMRs, there is unprecedented opportunity to uncover temporal disease correlations which have not previously been recogniz.
A computational method was recently applied to study temporal disease trajectories among chronic diseases using the Danish National Patient Registry.2 The algorithmic foundation of this method (the Danish method) is based on a key computational approach for identifying temporal correlations among diseases. Briefly, the Danish method utilizes the following principle: For every disease pair of X and Y, the exposed group is defined as all patients with diagnosis of disease X. For each patient in the exposed group, a set of gender, ethnicity and age-matched control subjects without disease X is selected to form the corresponding comparison group. The number of case subjects, with diagnosis of disease Y, is counted in both the exposed and the comparison groups. Then, by applying the binomial statistical test, the Danish method examines whether the proportion of case subjects is significantly higher in the exposed group than in the comparison group. If higher, the Danish method determines that the earlier diagnosis of disease X may have a significant temporal correlation with the later diagnosis of disease Y. The results of this approach provide the foundation for subsequent analysis of connecting significantly correlated individual disease pairs into trajectories consisting of multiple diseases.
Although the Danish method is a significant step toward uncovering unknown disease correlations, there are several inherent limitations to the approach. First, the amount of lag time between the onsets of different diseases is not characterized. For example, if a patient in the exposed group with disease X developed disease Y after 1 year, and another patient in the comparison group without disease X developed disease Y after 4 years, the Danish method would treat the two cases identically. This limitation occurs as a result of the Danish method’s categorical designation of the presence of disease Y within a 5-year window, rather than as a continuous time-to-event variable. Second, confounding factors cannot be easily accounted for by the Danish method as a result of simple stratifications. To account for the effects of gender and ethnicity, the Danish method requires identical gender and ethnicity between patients in the exposed and comparison groups. Such stratification may be plausible for categorical data (eg, gender and ethnicity), but is not suitable for continuous variables (eg, age, blood pressure, body mass index, etc.) which often cannot be stratified into workable categories. Finally, no readily available software package of the Danish method has been provided for use by other investigators.
Our approach aims to overcome the limitations of the Danish method by integrating two independent complementary approaches—Cox Proportional Hazard (Cox-PH) regression and Random Forest (RF) survival analysis. These approaches treat time to onset of disease as a continuous variable, and flexibly incorporate categorical and continuous covariates into the modeling of temporal relationships between disease pairs. The output of disease correlations may be further analyzed using our customized visualization tool.