gms | German Medical Science

Background: In regional negotiations between the statutory health insurance companies and the statutory health insurance physicians in Schleswig-Holstein the MRG system was chosen as the primary tool for the controlling of prescribed drugs in 2017. In two other regions there are evaluations of the system. In [3,8,9] the concept of Morbidity Related Group (MRG) was introduced in order to get a main drug prescription class for patients with respect to a physician and quarter. It was introduced in analogy to the Diagnose Related Group (DRG) in the hospital setting based mainly on diagnoses. Prescription analysis can utilize the five different resolution levels of the international anatomic-therapeutic-chemical (ATC) classification system. Interaction effects and treatment intensities and changes are chronologically interconnected by using prescription dates. Within that ATC framework the patient level is of minor importance. In the MRG setting we look for the group with the highest drug costs on the third ATC level (for characters) within a quarter for each consul- ted physician for a certain patient. This group should strongly be related to the morbidity of the patient and therefore it is called Morbidity Related Group. Thereby one considers the costs as a proxy for the severity of drug treatment.

Materials and Methods: We utilized prescription and diagnostic data of the most northern federal state of Germany (Schleswig-Holstein) from quarters 3/2015 till 2/2016. The analysis is related to patients, quarters and physicians. That means, that a patient is counted as much as pairs of quarters and physicians appear. There are 8.645 Million patients in the drug prescription data and 11.117 Million patients in the ICD-10 data. The C-related programming language awk is used for the computations. The visualization was done in Mathematica by Wolfram Reasearch and Microsoft Excel. As stated in the introduction, the basic MRG is determined by the ATC3 (four characters) with the highest costs with respect to patient, quarter and physician using prescription data. Thus, only patients with drug prescriptions can get a MRG. In analogy to the DRG system in inpatient care the basis MRG is extended by a degree of severity determined by age, multimorbidity (measured by polypharmacy) and prescription intensity. Hence, relations between MRG and ICD-10 codes with respect to multimorbidity are of interest. In the first step we consider patients with one ATC and one ICD-10 only. The resulting pairs provide ordered lists of ICD-10 per MRG and vice versa. Although the vast majority of drugs is prescribed in the field of multimorbid patients, we can use the obtained lists for additional considerations regarding all patients.

Let r(a) = (r_1(a); r_2(a),…,r_n(a)) be a vector with components that are given by the fraction of patients with age a and MRG I (i = 1,… ,n) where n is given by the number of MRG ordered for instance lexicographically. One can consider this with or without a gender restriction. For the age values a and b we consider the Manhattan distance. We can consider an inversion problem. If there is given a vector of disease fractions s we want to determine the respective age by a minimum problem. A vector of a certain subgroup of patients with a certain social status or insurance type with given age may optimally match a vector of a different age group. This can be interpreted as a higher or lower biological age. It has been already remarked, that polypharmacy is one factor for the determination of severity levels. An alternative model can be built applying polypharmacy instead of MRG. We consider an age dependent polypharmacy vector v(a) where the component v_i(a) describes the fraction of patients with i different drug groups (ATC3). Again, a Manhattan distance can be defined as usual minimum problem with a related inversion problem. Raising the question if the MRG-based or the polypharmacy-based model is more suitable for determination of the biological age of any chosen subgroup.

Results: For three example MRGs those diagnoses having a higher conditional probability then in the unconditional case are listed. Within the basis MRG M01A (Antiinflammatory and antirheumatic products) 33.0 % of the patients have ”dorsalgia“ (M54). In an age and gender adjusted patient group without the condition M54 only 17.8 % have a M54 diagnosis. Without age and gender adjustment we get 18.2 % (p3). In patients with MRG M01A (anti-inflammatory and antirheumatic drugs) only three diagnoses resulted in an increase in the conditional probability. For betablocking agents the same hold true for six ICD-10. The top ranking diagnosis is I10 (”essential primary hypertension“). The most significant diagnosis for patients with antidepressants is a F32 (”depressive episode“). For each example MRG the model has determined corresponding a top level diagnosis, for MRG M01A (Antiinflammatory and antirheumatic producs) the diagnosis M54 (”dorsalgia“), for MRG C07A (Beta blocking agents) the diagnosis I10 (”essential primary hypertension“) and for MRG N06A (Antidepressants) the diagnosis F32 (”depressive episode“). Generally the age dependent vectors r(a) determine the subgroup with corresponding age a. Age distances at least locally, but also for the age between 25 an 95 years, the Manhattan distance increases monotonically (independence of b while fixing a or vise versa respectively). In large areas we detect age sharply. This allows for comparison of models with and without gender components.

Conclusion: The MRG determines an unique type of patient based on drug prescription data labeled by a drug group. Furthermore we can construct another unique type using the number of prescribed drug groups (polypharmacy). If we want to analyze diagnosis structures in relation to age, gender, geographical regions or social status, a well defined patient type might be useful. Reversely, starting with a special diagnosis and asking for the probability of getting a special drug or drug group (again with age and gender standardization) might be of interest. The presented modeling approach can be applied in both directions. This flexibility offers a wide range of applications especially when patient orientation is necessary for the development of new forms of care. The need of an individualized medicine in certain patient subgroup can also be met and synchronized with the present risk adjustment scheme in the German statutory health insurance. This risk adjustment scheme might also profit by redefinition of patient groups and underlying parameters.