Key words Routine data - DIMDI - type 2 diabetes - early benefit assessment - target population
Schlüsselwörter Routinedaten - DIMDI - Diabetes mellitus Typ 2 - frühe Nutzenbewertung - Zielpopulation
Introduction
Since 2011, an early benefit assessment has been required for all new drugs launched
in Germany [1 ]. Evidence submitted in dossiers
by pharmaceutical companies is assessed by the Institute for Quality and Efficiency
in Health Care (IQWiG) Germany’s health technology agency, followed by an
appraisal by the supreme decision-making body in the German health care system, the
Federal Joint Committee (G-BA). The exact determination of the size of the
appropriate target population (i. e. patients eligible for a specific drug
within a one-year period) is important for subsequent price negotiations. The target
population in type 2 diabetes is classified into several groups based on the type of
antidiabetic medication prescribed. As a result, patients are assigned to the
following, most common, groups: patients with antidiabetic monotherapy, with dual
fixed- or loose-dose combination therapy, and with triple or more combination
therapy. If insulin is part of the therapy regimen, classification of patients took
place separately.
In type 2 diabetes, the size of the target population varies considerably between the
company dossiers submitted for assessment [2 ].
In Germany, non-commercial public health surveillance databases are currently only
in the process of being established [3 ].
Pharmaceutical companies therefore usually obtain information on the size of the
target population reported in dossiers from routine data analyses of various
commercial databases. These analyses differ with regard to the database, the
inclusion criteria and the calculation methods applied. Furthermore, the reporting
of these methods is often incomplete [2 ].
Our aim was to explore whether routine data from all persons insured in German
statutory health insurance (SHI) funds can be used to derive information on the size
of the target population in type 2 diabetes. So far, no standard procedures are
available to define different types of diabetes patients with the explicit use of
routine data [4 ]. Reported possible
misclassifications of type1 and type 2 diabetes are the main challenge [5 ]. As an indicator of the validity of our
methodology, we also decided to calculate the prevalence of type 1 and 2 diabetes in
the German population to enable a comparison with published data. In this
publication, we focus on the methodology of the approach. More information about the
results, as compared to other approaches, will be available in subsequent
publications as well as in the final report on the IQWiG website.
Methods
The prevalence of diabetes varies between different SHI funds in Germany due to
differences in the characteristics of the SHI members (e. g. distribution of
age, sex, social status, regions). We therefore decided to use data from the German
Institute of Medical Documentation and Information (DIMDI), which has a database
covering all of the approximately 70 million SHI fund members [6 ]. The DaTraV
(“Datentransparenzverordnung”, data transparency regulation) data
are generated from the yearly national morbidity risk compensation scheme
(Morbi-RSA) between the different SHI funds and are collected by the Federal
Insurance Office. Since the implementation of the data transparency regulation in
2012, authorized organizations may use aggregated data to answer predefined research
questions. In this process, to guarantee the anonymity of the SHI members, a group
reported in the result tables has to consist of at least 30 patients. The DIMDI
DaTraV database contains, among other things, the following information [7 ]: administrative data of the SHI members
(e. g. year of birth, sex, days insured), inpatient and outpatient diagnoses
(International Classification of Diseases ICD-10 code), data on outpatient
medication (pharmaceutical registration number). No information on specific medical
parameters (e. g. HbA 1c) is available. Data on SHI members whose membership
terminated (for whatever reason, e. g. death) are eliminated from the
database. At the time of the design of our analysis (2017), data of the year 2013
were the most recent available at DIMDI.
A comprehensive database query is necessary to access the DaTraV data. The query is
developed by the applicant or by the DaTraV data analyst. For the generation of the
database query and the analysis of the results, an in-depth understanding of routine
data analyses and of diabetes and its treatment is required. The project team, which
was formed at the beginning of 2017, comprised a diabetologist, an expert for
routine data analyses of patient data (member of the research group “PMV
forschungsgruppe”), and IQWiG researchers with expertise in early benefit
assessments of antidiabetic drugs. Based on the structure of the DIMDI DaTraV
database, the team had to define selection criteria in order to
(1) Reliably identify patients with diabetes,
(2) Distinguish patients with type 2 diabetes from patients with other types of
diabetes and
(3) Classify patients into different medication groups.
Moreover, the target population referred to patients with type 2 diabetes
continuously treated within a period of exactly one year.
In several project meetings, we discussed different approaches regarding the choice
and combination of selection criteria approaches like the one applied in the
CoDiM-study analyzing the cost burden of diabetes mellitus [8 ]
[9 ]. A data-based concept was chosen: Before
finalizing the database query, we repeatedly tested our set of selection criteria in
the database of a large SHI fund and examined the impact of modifications.
Results
1. Identification of patients with diabetes
As the target population had to have been continuously treated over a one-year
period, only those patients insured continuously in 2013 and at least one day in
2014 were included. Moreover, detailed and consistent data on the age and sex of
these patients had to be available in the database. To identify all patients
with diabetes, we chose a combination of three criteria: a corresponding ICD
diagnosis, repeated prescriptions of antidiabetic drugs, and exclusion
criteria.
All inpatients and outpatients for whom a repeated ICD diagnosis for diabetes
(acc. to E10 to E14, [Table 1 ]) could be
found in at least two quarters of 2013 were included (so called M2Q criteria).
For all outpatients, a specification of the diabetes diagnosis
“G” (“gesichert”), i.e
“confirmed” was essential to exclude those patients for whom the
diagnosis had not been conclusively confirmed during the period considered.
Table 1 ICD-10 diagnoses.
ICD-10 code
Description
E10.-
Type 1 diabetes mellitus
E11.-
Type 2 diabetes mellitus
E12.-
Malnutrition-related diabetes mellitus
E13.-
Other specified diabetes mellitus
E14.-
Unspecified diabetes mellitus
In cases where a corresponding ICD diagnosis for diabetes existed in only one
quarter, we also considered prescriptions of antidiabetic drugs. To identify
these prescriptions, at first, the pharmaceutical registration numbers in the
database had to be linked to corresponding Anatomic-Therapeutic Chemical (ATC)
classification system codes [10 ]). All ATC
codes for antidiabetic drugs (A10) were included, with a separate consideration
of insulin (A10A) and other antidiabetic drugs (A10B). The ATC code A10X (other
antidiabetic drugs) was excluded, as the number of cases was too small to ensure
the anonymity of the SHI members and no information could therefore be obtained
from the DIMDI DaTraV database. In cases of no existing diabetes diagnosis, we
included those cases with at least two days of prescriptions of antidiabetic
drugs within the one-year period and no documentation of other specific
diagnoses (ICD-10 diagnoses leading to exclusion are listed in [Table 2 ]).
Table 2 ICD-10 diagnoses excluded.
ICD-10 code
Description
E28.2
polycystic ovarian syndrome
E66.-
obesity
O24.4 / O24.9
diabetes mellitus arising in pregnancy
R73.-
elevated blood glucose level
[Table 3 ] presents an overview of the
combination of the three inclusion criteria.
Table 3 Combination of the inclusion criteria.
Criteria
Included
≥2 quarters with a diagnosis of diabetes mellitus
1 quarter with a diagnosis of diabetes mellitus
AND ≥1 day with a prescription of an
antidiabetic drug
≥2 days with a prescription of an antidiabetic drug
WITHOUT an exclusion diagnosis*
Excluded
1 day with a prescription of an antidiabetic drug
WITHOUT a diagnosis of diabetes mellitus
≥2 days of prescriptions of an antidiabetic drug
WITHOUT a diagnosis of diabetes mellitus but
WITH an exclusion diagnosis*
1 quarter with a diagnosis of diabetes mellitus
WITHOUT a prescription of an antidiabetic drug
*see [Table 2 ]
2. Classification of patients with type 2 diabetes
To identify all patients with type 2 diabetes as accurately as possible, we
developed a methodological approach in cooperation with our diabetologist. Since
patients could have visited different physicians in 2013 (e. g. general
practitioner, diabetes specialist, hospital physicians) who made different
diagnoses, it is possible that different ICD-10 diagnoses were available for the
same patient within the course of the year analysed. To distinguish between
patients with different types of diabetes, we developed a classification scheme
based on a combination of ICD and ATC codes and particularly considered
prescriptions of insulin and other antidiabetic drugs ([Fig. 1 ]).
Fig. 1 Flowchart for the classification of patients according to
their type of diabetes. ADT = other antidiabetic drugs than
insulin.
To classify patients as having type 2 diabetes, the following constellations
needed to apply:
an ICD-10 diagnosis of E11.- or E12.- and no ICD-10 diagnosis of
E10.-
a combination of the ICD-10 diagnosis E10.- and E11.-/E12.- and
prescriptions of other antidiabetic drugs than insulin (insulin
prescriptions were also possible)
a combination of the ICD-10 diagnosis E10.- and E11.-/E12.- and
no antidiabetic prescriptions
an ICD-10 diagnosis of E10.- and no insulin prescriptions
an ICD-10 diagnosis of E14.- alone and prescriptions of other
antidiabetic drugs than insulin (insulin prescriptions were also
possible)
an ICD-10 diagnosis of E14.- alone and no prescription of an antidiabetic
drug
patients without the ICD-10 diagnoses E10.- to E14.- but with
prescriptions of other antidiabetic drugs than insulin (insulin
prescriptions were also possible)
To classify patients as having type 1 diabetes, the following constellations
needed to apply:
An ICD-10 diagnosis of E10.- and no ICD-10 diagnosis of E11.- or E12.-
and at least one prescription of insulin
A combination of the ICD-10 diagnoses E10.- and E11.-/E12.- and
prescriptions of insulin without prescriptions of other antidiabetic
drugs.
Patients who had an ICD-10 diagnosis of E13.- without an ICD-10 diagnosis of
E10.- or E11.- were classified as “type 3 diabetes” (other
specified type) and patients who had an ICD-10 diagnosis of E14.- without an
ICD-10 diagnosis of E10.- to E13.- and with prescriptions of insulin without
prescriptions of other antidiabetic drugs were classified as having an
“unspecified type of diabetes”.
3. Classification of patients into medication groups
In early benefit assessments, the target population in type 2 diabetes consists
of the following groups:
Patients with antidiabetic monotherapy (e. g. metformin)
Dual fixed- or loose-dose combination therapy (e. g.
metformin+another oral antidiabetic )
Triple or more combination therapy.
If insulin is part of the therapy regimen, a separate classification of patients
takes place:
Patients treated continuously with one antidiabetic drug in the course of the
reference year 2013 can simply be classified as patients receiving monotherapy.
However, difficulties arise with the documentation of prescriptions of 2 (or
more) different antidiabetic drugs within the reference year. As packages of
antidiabetic drugs can contain different numbers of tablets with different
dosages (and different dosages can be prescribed for different patients),
packages can last for different treatment periods. In combination therapy,
recurrent prescriptions are therefore not necessarily issued on the same day. In
this case, it is difficult to distinguish between patients with a dual
combination therapy and patients with a trial therapy (e. g. the drug
was changed immediately after initiation because of intolerance) or patients
with a switch from one monotherapy to another after a longer period of time. We
therefore decided to take the individual last date of the prescription of an
antidiabetic drug and then look for a previous prescription within the same
group of drugs ([Table 4 ]) in the course
of the reference year (validation of prescriptions). With the documentation of
another prescription of an antidiabetic drug that belonged to the same group of
drugs in this period, the patient was classified in the therapeutic group
according to his/her medication. If no further prescription of an
antidiabetic drug belonging to the same group of drugs existed in the given
period, the patient was not included. By this method of validation of the
prescriptions, we only included continuously treated patients. In respect of
insulin, we separately considered treatment with basal, bolus and inhaled
insulin as well as the fixed-dose combination of basal and bolus insulin
(premixed insulin). By doing this, we were able to distinguish between different
treatment regimens such as conventional therapy (CT – fixed-dose
combination of basal and bolus insulin) and intensified conventional therapy
(ICT – loose-dose combination of basal and bolus insulin).
Table 4 Groups of anti-diabetic drugs that are considered in
the approach of the individual last prescription.
Group of drug
ATC codes
Insulins
A10A-
Bolus insulin
A10AB
Basal insulin
A10AC, A10AE
Premixed insulin
A10AD
Inhaled insulin
A10AF
Metformin
A10BA02-A10BD03, − 05,
− 07,− 08,
− 10
Sulfonylureas
A10BB-A10BD04, − 06, − 15
DPP-4 inhibitors
A10BH-A10BD07, − 08, − 10
Glinides
A10BX02, − 03
SGLT2 inhibitors
A10BX09
Thiazolidinediones
A10BG02, − 03 A10BD03 to 06
GLP-1 receptor agonists
A10BX04, −07, −10
Other antidiabetic drugs
A10BF01, −02
To determine the appropriate period for validation, we chose a data-based
approach: Within our test database, we analysed different periods (from 30 to
365 days before the last prescription). The median period was 90 to 100 days
before the last individual prescription. Within a 90-day period, up to
75% of the recurrent prescription cases would be covered; in a 180-day
period, it would be over 90%. The results of the 365-day period showed
little difference from results of the 180-day period. To cover 100% of
all cases with a recurrent prescription, a 2-year period would be necessary. We
decided to work with the 180-day period, as it covered the vast majority of
recurrent prescriptions within one year. This allowed us to use data solely from
the reference year 2013, thus avoiding inconsistencies of methodological
approaches.
By proceeding this way, we classified patients with continuous prescriptions of
one group of antidiabetic drugs as patients with monotherapy, while patients
with continuous prescriptions of 2 groups of antidiabetic drugs received the
classification as patients with dual combination therapy, and so forth.
Data processing at DIMDI
Based on the above considerations, the expert for routine data analyses (PMV)
developed the database query, which was subsequently submitted to DIMDI. Both
project partners, IQWiG and PMV, made use of the possibility to apply as
partners. This allowed us to share technical information between all involved
institutions (IQWiG, PMV, DIMDI) during the data processing procedure. As only
information on the drug ID code (“Pharmazentralnummer”, PZN) is
available in the DIMDI DaTraV database, we obtained permission from the Research
Institute of the Local Health Care Funds (WidO) to use their main drug database
of the year 2013, containing information on the relationship between the ATC and
PZN codes. DIMDI screened the resulting tables created by the analysis scripts
(written in Structured Query Language, SQL) to ensure the anonymity of the SHI
members. As mentioned previously, no results were available for groups with
fewer than 30 cases in at least one cell of the tables. The project partners
(IQWiG and PMV) checked the preliminary results twice at the DIMDI guest
workstation. Through this review, it was possible to assess whether an
additional aggregation of the data due to data protection issues and revision of
the script was required.
Discussion
The main aim of our study was to explore whether routine data from German statutory
SHI funds is a sufficient basis to derive information on the size of the target
population in type 2 diabetes, while excluding patients with other types of diabetes
and categorizing those with type 2 diabetes according to their medication. Moreover,
the study was conducted to obtain deeper insight into the specific methodological
challenges and their impact on the size of the target population.
In contrast to other diabetes studies, which focus primarily on the determination of
the prevalence and incidence of diabetes mellitus (e. g. [11 ]
[12 ]) we considered not only the diagnosis,
but also the antidiabetic medication of the patients as selection criteria. By
combining ICD diagnoses and medication, we tried to detect as many patients with
diabetes as possible. To reduce misclassifications, some indications were regarded
as exclusion criteria, namely indications in which patients do not seem to suffer
from diabetes, but still receive an antidiabetic medication. Based on their
antidiabetic therapy we also classified some patients as having diabetes even if
they did not have a corresponding ICD diagnosis. Even under considering the
exclusion criteria, this could still be a source of overestimation of patient
numbers.
It was possible to develop a methodological approach that allowed categorizing
patients with type 2 diabetes into respective medication groups according to their
prescriptions. For the target population, only patients with type 2 diabetes and
continuous therapy were eligible. Moreover, the target population referred to
exactly one year. We therefore decided to use only information in the database from
the year 2013. This approach implies that persons who terminated their SHI
membership in 2013 are not included in our dataset, as automatic deletion of such
data from the DIMDI DaTraV database takes place in the year of termination. To
obtain information on the number and treatment of these persons in the year 2013,
information from the previous year 2012 would have been required. We decided against
this approach, as in 2012 changes to the ATC classification were made, which would
have led to methodological problems. Moreover, the classification of patients into
the medication groups (individual last prescription and another prescription within
the same group of drugs within 180 days before the last prescription) only refers to
a one-year period and does not consider information from the previous year.
The fact that SHI members with a terminated membership in the year analysed were not
included in the medication groups implies that there will be some underestimation of
the size of the target population. The reason for this is that the prevalence of
diabetes might have been higher in these patients than in the remaining SHI members.
An underestimation could also occur from our approach to take the individual last
prescription and look for another prescription within the same group of drugs within
the course of the year. This implies exclusion of incident patients diagnosed in the
last quarter, as no further diagnosis or prescription of an antidiabetic drug was
available. In total, the results on the numbers of patients in the medication groups
were largely consistent with the findings in company dossiers submitted to the G-BA
[2 ] . According to our diabetologist, they
are plausible from a medical point of view.
Compared to other routine data sources, the DIMDI DaTraV database has the advantage
of including information on all SHI fund members in Germany [6 ]. No information is lost when persons switch
from one SHI fund to another. Moreover, no adjustments of the results are necessary
to ensure appropriate distributions of key parameters such as age and sex. This
would have been necessary with the use of data from only one SHI fund.
Limitations arise from the fact that pharmaceutical companies have no authorization
to access the DaTraV data: they cannot use the DaTraV data for early benefit
assessments on their own. Further limitations relating to most types of routine data
analyses apply: Without further detailed medical information, only diagnosed
patients can be identified, while no information on the estimated number of unknown
cases with diabetes is available. Moreover, the data on ICD diagnoses, medications,
dates for doctor’s appointments and prescriptions can contain mistakes.
These arise, among other things, from coding procedures and different standards for
defining the type of diabetes. Results based on the DIMDI DaTraV database therefore
reflect more real-world situations. We tried to address these problems by
implementing validation steps; however, it is still possible that some patients were
wrongly included in our dataset.
Generation of the DIMDI DaTraV database is based on the data of the Morbi-RSA and
contains only a selection of SHI data [6 ]. It
therefore contains information more relevant to costs than to medical aspects.
Information about inpatient and outpatient care is restricted as, for example, no
information is available about compensations concerning the so called German Uniform
Assessment Standard (“Einheitlicher Bewertungsmaßstab”, EBM)
or Diagnosis Related Groups (DRG) [6 ].
Nevertheless, the information on the indication of diabetes available in the
database was sufficient to determine the target population. Further limitations can
arise with the use of the DIMDI DaTraV database for other indications: Because of
the restricted actuality, the database is more suited to derive information about
the target population in indications with a relatively constant incidence and
prevalence over the years. For some indications, a more detailed medical information
(e. g. blood parameters, histology, stages of disease) or accounting
information (EBM or DRG numbers) could be necessary. Therefore, the usefulness of
the DIMDI DaTraV database to determine the target population in early benefit
assessments is restricted only to certain indications.
In summary, DIMDI data are a valuable source for obtaining information on the target
population in type 2 diabetes in early benefit assessments. The methodological
approach seems to be suited to determine the target population in type 2 diabetes.
For the comparison of routine data analyses within the same indication, a detailed
and standardized description of the methodological approach is important [13 ].
