Man vs. machine: comparison of pharmacogenetic expert counselling with a clinical medication support system in a study with 200 genotyped patients

Background Medication problems such as strong side effects or inefficacy occur frequently. At our university hospital, a consultation group of specialists takes care of patients suffering from medication problems. Nevertheless, the counselling of poly-treated patients is complex, as it requires the consideration of a large network of interactions between drugs and their targets, their metabolizing enzymes, and their transporters, etc. Purpose This study aims to check whether a score-based decision-support system (1) reduces the time and effort and (2) suggests solutions at the same quality level. Patients and methods A total of 200 multimorbid, poly-treated patients with medication problems were included. All patients were considered twice: manually, as clinically established, and using the Drug-PIN decision-support system. Besides diagnoses, lab data (kidney, liver), phenotype (age, gender, BMI, habits), and genotype (genetic variants with actionable clinical evidence I or IIa) were considered, to eliminate potentially inappropriate medications and to select individually favourable drugs from existing medication classes. The algorithm is connected to automatically updated knowledge resources to provide reproducible up-to-date decision support. Results The average turnaround time for manual poly-therapy counselling per patient ranges from 3 to 6 working hours, while it can be reduced to ten minutes using Drug-PIN. At the same time, the results of the novel computerized approach coincide with the manual approach at a level of > 90%. The holistic medication score can be used to find favourable drugs within a class of drugs and also to judge the severity of medication problems, to identify critical cases early and automatically. Conclusion With the computerized version of this approach, it became possible to score all combinations of all alternative drugs from each class of drugs administered (“personalized medication landscape “) and to identify critical patients even before problems are reported (“medication alert”). Careful comparison of manual and score-based results shows that the incomplete manual consideration of genetic specialties and pharmacokinetic conflicts is responsible for most of the (minor) deviations between the two approaches. The meaning of the reduction of working time for experts by about 2 orders of magnitude should not be underestimated, as it enables practical application of personalized medicine in clinical routine.