Comparison of Different Implementations of a Process Limiting
Pharmaceutical Expenditures Required by German Law
Timo Emcke
1
, Thomas Ostermann
2
, Marc Heidbreder
3
and Reinhard Schuster
4
1
Chair of Department of Prescription Analysis, Association of Statutory Health Insurance Physicians,
Bismarckallee 1-6, 23812 Bad Segeberg, Germany
2
Chair of Research Methodology and Statistics in Psychology, Witten/Herdecke University,
Alfred-Herrhausen-Straße 50, 58448 Witten, Germany
3
Medical Advisory board of Statutory Health Insurance in Northern Germany, Katharinenstraße 11,
23554 L
¨
ubeck, Germany
4
Chair of Department of Health Economics, Epidemiology and Medical Informatics, Medical Advisory Board of Statutory
Health Insurance in Northern Germany, Katharinenstraße 11, 23554 L
¨
ubeck, Germany
Keywords:
Controlling Measures, Outpatient Care, Morbidity Adjusted Benchmark, Prescription Data, MRG, Big Data.
Abstract:
German legislation demands controlling measures for outpatient drug costs. As of 2017 the health insurance
actors rose to a challenge to reform the benchmark system on the federal state level. We look at the previous
system applied until 2015, the improvements in 2016 and the method the regional parties agree on for 2017.
After discussing hard- and software systems and the underlying data briefly we describe the flaws of the old
approach and develop a general model for controlling measures in the outpatient field. Finally we present the
first real world applications of the new model: a patient type classification system leading to target costs and a
derived distance structure of physicians regarding their prescription behaviour.
1 INTRODUCTION
In Europe and especially in Germany rising pharma-
ceutical expenditures put the health service at risk.
Every modern health care system has to ensure the
quality and equity of care while keeping the cost
down. Therefore controlling measures were establis-
hed by the German legislation as early as in 1993.
Since 1995 this is subject to regional negotiations bet-
ween Statutory Health Insurances (SHI) and SHI as-
sociated physicians. This type of regulation aims to
limit expenditures per patient without restricting the
necessary treatment.
Of the exiting two types of instruments, the first
one puts German patients/cases in certain more or less
morbidity related cost groups, the other promotes or
restricts drug classes with different economic charac-
teristics but same curative effects. We will look at tho-
se using health insurance data of the German Federal
State Schleswig-Holstein in 2015.
In the years from 1995 till 2015 physician groups
got three different treatment case budgets for each
insurance status defined by statutory health insuran-
ce (member [M], dependent coverage [D] and retired
[R]). Some regions merged status [M] and [D]. Se-
veral expensive drug substances and pharmaceuticals
regulated by treatment regimen are excluded resulting
in internal inconsistencies and uncertainties regarding
all participating players.
Budgets are calculated using expenditure shares
for the mentioned case groups per physician group in
a reference period (last year) and the negotiated target
volume of expenditure for the resent year.
In December 2013 the social welfare court
of Dresden passed the sentence that guide va-
lues/budgets have to be based on age groups. Addi-
tionally the Federal Social Court judged that authori-
ties have an obligation to regulate atypical drug pres-
criptions. As an immediate consequence regarding the
budget calculation for 2016 four age groups superse-
ded insurance status: 0-15, 16-49, 50-64 and 65 and
above. Those groups, utilized in all statutory health
insurances, have a very poor age resolution for this
field of application in general.
From 2017 on, the federal legislator made re-
gional negotiated far-reaching reforms of control-
ling measures possible (Versorgungsst
¨
arkungsgesetz
= Supply Support Act). A new system developed in
this context is expenditure controlling by Morbidi-
ty Related Groups (MRG). MRG is an adaption of
Emcke T., Ostermann T., Heidbreder M. and Schuster R.
Comparison of Different Implementations of a Process Limiting Pharmaceutical Expenditures Required by German Law.
DOI: 10.5220/0006114800350040
In Proceedings of the 10th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2017), pages 35-40
ISBN: 978-989-758-213-4
Copyright
c
2017 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
35
the Diagnosis-Related-Group-System (DRG) used for
classification and compensation of hospital cases and
put into effect in 2003 by German legislation. It is ba-
sed on similar systems elsewhere: Since the first use
for hospital payment in the United States in the ear-
ly 1980s, DRG-type systems have become the main
method of hospital payment in the majority of OECD
countries. The German version (G-DRG) is based on
the Australian DRG-system (AR-DRG).
Hereinafter we will compare the systems based on
insurance status, age groups and MRG, including so-
me new results for MRG.
2 MATERIAL AND METHODS
For comparing the previous and the new controlling
measures we analyze detailed prescription data of the
Statutory Health Insurance in Schleswig-Holstein of
quarter two in 2015. There’s no benefit using annu-
al data due to the stable prescription framework. The
data on the prescription level are combined with ma-
ster data containing drug classes (ATC [anatomic-
therapeutic-chemical] with some additions for pro-
ducts not classified), physician groups and drugs to
be excluded. Treatment cases of the Association of
Statutory Health Insurance Physicians are also ad-
ded. Obtaining results required the processing of large
amounts of data.
The hardware used is a dedicated Debian
GNU/Linux Server administered by the Medical Ad-
visory board of Statutory Health Insurance in Nor-
thern Germany also used to generate consultation ma-
terials from the same data.
It runs a LAMP configuration (Debian
GNU/Linux, Apache 2.4, MYSQL Community
Edition 5.7 [extensive use of partitioning] and PHP 7
[with PEAR framework esp. for spreadsheet output]).
The inexpensive open source/free software setup
makes the cooperation of different administrative
bodies possible. The coding was done using the Perl
programming language.
The previous model used till 2016 applies pres-
cription data, treatment cases and status defined by
statutory health insurance/age groups. The implemen-
tation is straight forward. Treatment cases in a certain
age/status group get their share of the negotiated vo-
lume of expenditure based on the development of last
years‘expenditures and treatment cases.
The new MRG-model requires prescription data,
the ATC classification and physician group informa-
tion depending on the model configuration. It can be
defined as follows:
B = set of physicians/practices
F = set of physician groups
There is a transformation mapping physi-
cians/practices to groups: f = f (b) while splitting up
practices containing different physician groups.
P(b) = patients of b B, b = b(p) is the mapping
of patients and physicians whereas the transformation
D = D(p) maps patients p P(b) to the prescribed
drugs. Multiple prescriptions of one drug are coun-
ted repeatedly. o(d) is a quantity factor for d D
representing the ration of package size of the pres-
cription drug in relation to the biggest one available.
A pharmaceutical classification system (e.g. ATC4)
as transformation: a = a(p),a A used identificati-
on of similar medicinal products. The drugs d D
are linked to costs by the cost function: k = k(d),k
R,k > 0.The age of the patient is defined by: t = t(p)
in five-yearly stages. A MRG is a pair d = (c, s)
[c:basic MRG, s:degree of severity] with c A,s
Z,0 s 9.
Cost per ATC =
¯
k(p,a
) =
dD(p),a(p)=a
k(d)
ATC with the highest costs = basic MRG is characte-
rized by
¯
k(p,c)
¯
k(p,a) for all a A. In case of the
occurrence of several c
i
the lexicographically domina-
ting element is chosen. c = c(p) is the transformation
to determine patients basic MRG. Number of ATC4
groups per patient (multimedication) is defined as:
v(p) = #{a A :
¯
k(p,a) > 0}.
The number of prescriptions for patient p P assi-
gned to basic MRG c(p) is represented by:
¯o(p) = w
dD(p):a(d)=c(p)
o(d)
with
w(x) =
x, if x Z
bxc, if x / Z.
We define threshold values for subgroups:
(v
0
,··· ,v
9
) = (0.5, 0.75,1.25,1.5, 2.0,2.5, 5,10)
i(v) = i is true if v
i
v < v
i
+ 1. m(X) shall be the
mean of x X. The costs of basic MRG c A in the
physician group are defined as:
k
1
(c
, f
) = m
pP: f (b(p))= f
,c(p)=c
¯
k(p,c
)
and adding the age dimension the term changes to:
k
1
(c
, f
,t
) = m
pP: f (b(p))= f
,c(p)=c
,t(p)=t
¯
k(p, c
)
whereby the age related severity is given by:
i
1
(c, f ,t) = i
k
1
(c, f ,t)/k
(c, f )
.
Costs differentiated by multimorbidity are expressed
by the formula:
HEALTHINF 2017 - 10th International Conference on Health Informatics
36
k
2
(c
, f
, j
) = m
pP: f (b(p))= f
,c(p)=c
,v(p)= j
¯
k(p, c
)
with the corresponding degree of severity:
i
2
(c, f , j) = i
k
2
(c, f , j)/k
(c, f )
.
The same can be done by looking at prescription in-
tensity:
k
3
(c
, f
, j
) = m
pP: f (b(p))= f
,c(p)=c
, ¯o(p)= j
¯
k(p, c
)
i
3
(c, f , j) = i
k
3
(c, f , j)/k
(c, f )
.
Total degree of severity is given by:
i(p) = max
i
1
(c(p), f (b(p),t(p)),i
2
(c(p)), f (b(p)), v(p)),
i
3
(c(p)), f (b(p), ¯o(p))
.
The MRG including severity levels is recalculated
with respect to physician groups:
k
g
(c
, f
, j
) = m
pP: f (b(p))= f
,c(p)=c
,i(p)= j
¯
k(p, c
)
.
Thereby we get the target cost for benchmarking the
physician:
k
#
=
p=P(b)
k
g
c(p), f (b(p)),i(p)
.
In our setting we look for the group with the hig-
hest drug costs within a quarter for each consulted
physician for a certain patient. This group should
strongly be related to the morbidity of the patient
and we will call it therefore Morbidity Related Group
(MRG). One considers the costs as a proxy for the
severity of drug treatment and could also take other
weight functions instead of cost. The following is an
example regarding a diabetes patient who belongs to
the basic group A10A (Insulins and analogues) with
total patient cost of 1,536.75 e:
Table 1: Example of (basic) MRG determination.
cost nr ATC substance drug amount
320.74 1 B01AF01 Rivaroxaban XARELTO 15 mg 98
272.61 1 N06AX21 Duloxetine CYMBALTA 60 mg 98
248.02 2 A10AD04 Insulin Lispro LIPROLOG Mix 25 10X3
208.25 7 V04CA02 Glucose CONTOUR Test-
streifen
50
159.35 1 N02AA55 Oxycodone TARGIN 10 mg/5
mg
100
124.01 1 A10AD04 Insulin Lispro LIPROLOG Mix 50 10X3
112.35 1 N02AA55 Oxycodone TARGIN 5 mg/2.5
mg
100
23.97 1 C10AA01 Simvastatin SIMVA ARISTO 40
mg
100
19.22 1 C03CA04 Torasemide TORASEMID AL 20
mg
100
16.27 1 N02BB02 Metamizole
Sodium
NOVAMINSULFON
1A
100
15.41 1 H03AA01 Levothyroxine
Sodium
L-THYROX HE-
XAL 125
100
13.98 1 C07AB12 Nebivolol NEBIVOLOL Glen-
mark 5 mg
100
As an initial adjustment factor the age of patients
can be applied. In each 5 year group of patients the
ratio of costs per patient in the subgroup compared
to the whole MRG was considered. If the ratio lies in
certain intervals (0-0.5, 0.5-0.75, 0.75-1.25,..., 10) the
age severity level 0,1,...,9 were assigned. The same
can be conducted with respect to other factors corre-
lated with morbidity. By using subgroup structures a
risk adjustment can be accomplished. All of this has
not to be precise on the level of the patient but on
the physicians level. Regarding the considered MRG
A10A (Insulins and analogues) seven degrees of se-
verity in the range of 101.27 e up to 1,385.61 e re-
sulted:
Table 2: Example of severity levels of MRG A10A.
degree cost in Euro number of patients
2 101.27 21
3 273.68 60,634
4 517.87 16,840
5 707.95 20,904
6 995.74 2,085
7 1,385.61 1,954
We divide the (basic) MRG into several severi-
ty levels that will be analysed by Lorenz curves and
the corresponding Gini coefficients: Additionally the
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
0% 10% 20% 30% 40% 50% 60% 70% 80% 90%
costs in %
patients in %
Lorenz curve for MRG AL04A (immunosuppressants) - general practicioners
Figure 1: Lorenz curve of MRG AL04A.
Shannons entropy (
p
i
log(p
i
)) can be applied to
the patient type structure in each physicians group
with respect to the MRG basic groups:
Table 3: Shannon entropy per physician group (1).
enthropy number of patients physician group
3.9985 994,220 general practitioners
3.8156 95,054 paediatricians
3.5635 3,548 non-specialised internists
3.4234 4,991 nephrologists
3.1437 1,955 haematologist/oncologists
3.1149 3,285 cardiologists
2.9632 2,961 gastroenterologists
Comparison of Different Implementations of a Process Limiting Pharmaceutical Expenditures Required by German Law
37
Table 4: Shannon entropy per physician group (2).
enthropy number of patients physician group
2.7608 84,660 dermatologists
2.7552 16,155 surgeons
2.7327 40,045 neck nose ear physician
2.6253 91,868 gynaecologists
2.4939 37,199 urologists
2.2263 49,653 neurologists
2.1716 1,325 endocrinologists
2.1210 3,851 rheumatologists
2.0282 40,149 orthopaedic
1.6552 2,123 anaesthetists
1.4179 51,819 ophthalmologists
1.3517 26,392 pulmonologists
1.1722 7,003 psychiatrists
3 RESULTS
The application of the treatment case oriented approa-
ches over the last decades showed that these systems
are incapable of considering age and progress related
increase of prescription costs. Recent analysis of the
age distribution of treatment cases in each Statutory
Health Insurance status group shows that the applied
age groups might be too coarse and unsuited as the
insurance status for the morbidity related depiction of
prescription costs per patient:
0
50,000
100,000
150,000
200,000
250,000
300,000
0 10 20 30 40 50 60 70 80 90 100
treatment cases
age
age dependent treatment cases per insurance status
D
M
R
Figure 2: Age dependent number of treatment cases per ins-
urance status/age groups.
The high correlation of the results of the two
methods applied until 2016 confirms that shifting to
age groups on the physicians level had practically no
benefit:
y = 0.9505x + 0.0514
R² = 0.9784
0%
50%
100%
150%
200%
250%
0% 50% 100% 150% 200% 250%
benchmarking result age groups
benchmarking result health insurance status
benchmarking practices using health insurance status/age groups
Figure 3: Correlation of benchmarks using health insurance
status vs. age groups.
Hence, a new system based on MRG is introduced
in 2017. There is little correlation between the results
obtained by the previous and the new results on the
practitioners level. That‘s due to the fact that many
factors were disregarded in the past and inconsisten-
cies were compensated by manual intervention:
y = 1,74x - 0,7687
R² = 0,2571
0%
25%
50%
75%
100%
125%
150%
175%
200%
0% 25% 50% 75% 100% 125% 150% 175% 200%
benchmarking results of practices using the
appoach applied until 2016
benchmarking result of practices using the MRG approach used in 2017
correlation of benchmarking results at the practitioners' level
Figure 4: Correlation of benchmarks applying the previous
(until 2016) and the new approach (2017).
Sorting the practices in ascending order for all af-
fected groups due to their MRG benchmarking result
and comparing those to the outcomes of the older
system demonstrates the progress in model adaption
made:
Figure 5: Results of MRG vs. system based on treatment
cases in 2016 (each line one practice).
HEALTHINF 2017 - 10th International Conference on Health Informatics
38
In those new MRG models all patients of a cer-
tain practice are classified and a specific structure for
each practice is the result. As an example, we consider
a physician with 14.0 % of his patients in the MRG
A10A (Insulins and analogues) and 11.8 % patients in
the MRG V04C (other diagnostic agents = test strips
measuring glucose). In this group of general practi-
tioners (GP) patients in those groups only account for
3.8 % in these two groups. The physician can thereby
be identified as a diabetologist:
Table 5: Patient structure of a diabetologist.
nr MRG nr. cost per prop. prop. drug droup
pat. patient doc. group
1 A10A 193 463.12 14.0% 2.4% Insulins and analo-
gues
2 V04C 162 307.76 11.8% 1.4% Other diagnostic
agents
3 H03A 86 22.28 6.3% 4.5% Thyroid preparations
4 A10B 82 185.78 6.0% 2.7% Oral blood glucose
lowering drugs
5 A02B 73 60.91 5.3% 7.2% Drugs for peptic
ulcer and gastro-
oesophageal reflux
disease (gord)
6 B01A 53 366.21 3.9% 4.0% Antithrombotic
agents
1 J01D 22 31.70 1.6% 2.4% Other beta-lactam
antibacterials
19 C10A 17 58.30 1.2% 3.0% Cholesterol and tri-
glyceride reducer
20 N03A 16 275.63 1.2% 1.4% Antiepileptics
After the formation of groups for all practi-
ces/physicians one can compare the MRG distributi-
ons and values in each group:
Table 6: MRG patient shares of orthopaedics.
frac. cost MRG drug group
pat. p. pat.
42.9% 19.95 M01A Antiinflammatory and antirheumatic products,
non-steroids
13.6% 26.26 H02A Corticosteroids for systemic use, plain
12.4% 23.81 N02B Other analgesics and antipyretics
8.2% 134.66 M05B Drugs affecting bone structure and mineraliza-
tion
6.1% 123.78 N02A Opioids
3.6% 92.18 B01A Antithrombotic agents
3.3% 40.71 M03B Muscle relaxants, centrally acting agents
2.5% 33.75 A02B Drugs for peptic ulcer and gastro-oesophageal
reflux disease (gord)
1.1% 2,515.02 L04A Immunosuppressive agents
1.0% 304.63 L01B Antimetabolites
Regarding orthopedics we observe a patient type
structure, in which 42.9 % of all patients belong to
the MRG M01A (antiinflammatory and antirheuma-
tic products, non-steroids). The 10 leading positions
cover 94.6 % of the costs. Costs again depend mainly
on the medical discipline. In oncology average costs
per patient are 15,288.17 e in the MRG L04A (im-
munosuppressive agents including all the other drugs
for the patient) versus 2,515.02 e for orthopedics. In
urology the top ten positions with respect to the num-
ber of patients cover 83.6 % of the costs. In the case
of GP these costs are only 44.2 %:
Table 7: MRG patient shares of urologists.
frac. cost MRG drug group
pat. p. pat.
34.8% 44.67 G04C Drugs used in benign prostatic hypertrophy
16.7% 136.12 G04B Other urologicals, incl. antispasmodics
9.9% 19.84 J01M Quinolone antibacterials
6.8% 618.61 L02A Hormones and related agents
4.9% 30.42 J01X Other antibacterials
3.1% 33.47 J01D Other beta-lactam antibacterials
2.0% 154.56 G03B Androgens
1.9% 22.13 J01E Sulfonamides and trimethoprim
1.7% 27.71 D01A Antifungals for topical use
1.7% 4,122.10 L02B Antimetabolites
Table 8: MRG patient shares of general practitioners.
frac. cost MRG drug group
pat. p. pat.
7.3% 62.55 A02B Drugs for peptic ulcer and gastro-oesophageal
reflux disease (gord)
5.4% 42.04 C07A Beta blocking agents
5.1% 34.67 M01A Antiinflammatory and antirheumatic products,
non-steroids
4.7% 24.25 H03A Thyroid preparations
4.2% 185.14 R03A Adrenergics, inhalants
3.8% 316.40 B01A Antithrombotic agents
3.8% 124.59 C09D Angiotensin II antagonists, combinations
3.4% 85.10 C09C Angiotensin II antagonists, plain
3.4% 30.68 C09A Ace inhibitors, plain
3.2% 88.42 N06A Antidepressants
The MRG patient shares can be utilized to gene-
rate distance measures for the clustering of all practi-
ces/physicians. Let p
k
m
be the fraction of patients with
MRG m (m M) for the physician k (k P). With
respect to the medical discipline s (s S) and let q
s
m
be the respective fraction. Let r and s be such fracti-
ons for physicians or medical disciplines we can use
a Manhattan distance:
iM
|r
i
s
i
|
Alternatively we can apply spherical distances on the
n-dimensional sphere where n is the number of MRG
classes with respect to the points:
r
m
r
jM
r
2
j
and
s
m
r
jM
s
2
j
or
r
m
and
s
m
The spherical distances are differentiable with respect
to the components of r and s and thereby is more sui-
table for optimization procedures.
Comparison of Different Implementations of a Process Limiting Pharmaceutical Expenditures Required by German Law
39
We can define the discipline t S of a physician
k P by the value s S for which:
mM
|p
m
k
q
s
m
|
has a minimal value. The distance of a physician to a
group measures to which extent he is typical or not.
Extreme values may be a hint for the need for spe-
cial considerations. One can use cluster methods in
order to receive a classification of physicians without
the use of their medical discipline which is primarily
determined by admission law.
4 CONCLUSIONS
The 2016 switch from health insurance status to age
groups did not eliminate the flaws of the old bench-
marking/budget approach. New promising ideas on
the regional level like MRG have a huge potential still
to be researched and utilized. The necessary data is
provided, hard-/software and knowledge are availa-
ble. Steady change and especially new form of he-
alth care require adapting benchmarking systems on
a sound data and legal foundation. Therefore MRG
seems to be a highly suitable approach meeting the
criteria.
REFERENCES
Bratzke, B. Spies, K.-P. and Krebs, S. (2012). Morbi-
dit
¨
atskomponente bei Arznei- und Heilmittelbudgets
einf
¨
uhren. Deutscher
¨
Arztetag, Drucksache V I-37.
Breyer, F. and Ulrich, F. (2000). Gesundheitsausgaben, Al-
ter und medizinischer Fortschritt: Eine Regressions-
analyse. Jahrb
¨
ucher f
¨
ur National
¨
okonomie und Stati-
stik, 220:1–17.
Busse, R. Panteli, D. and Krebs, S. (2015). Arzneimittel-
versorgung in der GKV und 15 anderen europ
¨
aischen
Gesundheitssystemen: Ein systematischer Vergleich.
Working papers in health policy and management.
Universit
¨
atsverlag der TU Berlin, 11.
GKV Versorgungsst
¨
arkungsgesetz (GKV-VSG) (2015).
BGBl., I:1211.
Gottwald, N. (2015). Damoklesschwert Richtgr
¨
oßen-
pr
¨
ufung. Im Dschungel der 25-Prozent-Regel. HNO-
Nachrichten. Springer, 3.
Korzilius, N. (2015). Arzneimittelrichtgr
¨
oßen:
¨
Arzte sollen
angstfrei verordnen. Dtsch Arztebl, 112.
Schuster, R. (2015). Morbidity Related Groups (MRG) and
drug economic index - a new concept after the age of
Richtgr
¨
oßen benchmarks in Germany. GAA 2015.
Schuster, R. and von Arnstedt, E. (2012). Mediziner-
technischer Fortschritt und demografischer Wandel
bei den GKV-Ausgaben im Vertrag
¨
arztlichen Bereich.
GMDS 2011.
Schuster, R. and Voss, H. (2012). Mediziner-technischer
Fortschritt und demografischer Wandel bei den GKV-
Ausgaben im DRG-Bereich im Krankenhaus. GMDS
2012.
Schuster, R. Emcke, T. and Schuster, F. (2016). Ver-
gleich der Richtgr
¨
oßenauswertungen f
¨
ur Arzneimittel
im Vertrags
¨
arztlichen Bereich mit Statusgruppen und
Altersgruppen. HEC 2016.
Sinowatz, F. (2015). Ressourcenalokation in der haus
¨
arztli-
chen Versorgung: eine empirische Untersuchung. Dis-
sertation LMU M
¨
unchen: Medizinische Fakult
¨
at.
Spitzenverband der Krankenkassen Bund und Kassen
¨
arzt-
liche Bundesvereinigung (2015). Rahmenvorgaben
nach 84 Abs. 7 SGB V -Arzneimittel- f
¨
ur das Jahr
2015.
Wersborg, T. (2006). Morbidit
¨
atsbezogene Richtgr
¨
oßen
zur Steuerung einer bedarfsgerechten und wirtschaft-
lichen Arzneimittelversorgung innerhalb der gesetzli-
chen Krankenversicherung in Deutschland. Disserta-
tion LMU M
¨
unchen: Medizinische Fakult
¨
at.
HEALTHINF 2017 - 10th International Conference on Health Informatics
40