Keywords claims data - record linkage - cause of death - cancer registry - routine data
Schlüsselwörter Sekundärdaten - Datenabgleich - Todesursache - Krebsregister - Routinedaten
Introduction
With nearly 10 million deaths in 2020, cancer is a leading cause of death worldwide
[1 ]. Organized screening programs
such as mammography screening for the early detection of breast cancer aim to reduce
cancer mortality [2 ]
[3 ]
[4 ]. For the mortality evaluation of
screening programs, information on causes of death is needed. Longitudinal health
insurance claims data, including individual information on screening participation
and cancer diagnoses, represent a potential data source for mortality evaluations
but contain no information on causes of death. Therefore, this information needs to
be linked from other data sources such as official databases or cancer registries
[5 ]. Alternatively, claims-based
algorithms validated against official death certificate information might also be
used [6 ].
In Germany, a mammography screening program (MSP) has been implemented nationwide
since 2009 [7 ]. Whether the program
reduces breast cancer mortality is currently evaluated using health insurance claims
data, cancer registry data, and screening unit data [7 ]
[8 ]
[9 ]
[10 ]. The claims data are available from
two separate data sources, the German Pharmacoepidemiological Research Database
(GePaRD), including 25 million individuals from four health insurance funds, and the
BARMER Data Warehouse, including 12.6 million individuals from another insurance
fund [11 ]. Because both data sources
include no information on causes of death, this information needs to be linked from
cancer registries, the only institutions conducting such linkages for the German MSP
evaluation. So far, the proportion of successful cause-of-death linkages has only
been determined for a linkage between two insurance funds included in GePaRD and one
cancer registry [12 ]. The proportions
of successful linkages are still unclear for other insurance funds and cancer
registries.
The purposes of this work were i) to determine the proportions of successful
cause-of-death linkages between the BARMER health insurance fund and three federal
cancer registries and ii) to investigate whether linked proportions differ by
region, year, and age.
Methods
Data sources
BARMER claims data
About 90% of the German population is statutorily (i. e. non-privately)
health insured. Of these, almost 9 million people from all over Germany are
members of Germany’s second-largest statutory health insurance fund, BARMER.
For this study, we used BARMER claims data that cover about 10% of the
German population [13 ]
[14 ]. The considered claims data
were from 2006 to 2018 and contain information on sex, year of birth,
federal state of residence, dates of insurance entry and termination, and
the reason for insurance termination (termination due to death or other
reasons, such as a switch to another insurance fund or private health
insurance). Further data include information on in- and outpatient diagnoses
as well as medical procedures and services. The data contain no information
on cause of death.
Cancer registry data
The history and legal basis of cancer registries in Germany differ among the
16 German federal states. Beginning in 1926 with the first attempts of
cancer registration in the city-state of Hamburg, all federal states had
committed to the Federal Cancer Registry Data Act
(Bundeskrebsregisterdatengesetz, BKRG, [15 ]) by 2009. They established
individual federal state laws, including quality and data management
guidelines, resulting in nationwide coverage with 15 epidemiological cancer
registries (Berlin and Brandenburg share their registries) focusing on
population-based outcomes and clinical cancer registries focusing on
clinical outcomes [16 ]
[17 ]. All German cancer
registries have official death certificate information for registered cancer
cases, irrespective of whether they died of cancer or not. The cancer
registry in North Rhine-Westphalia (NRW) has additionally full access to
population data, including causes of death for the whole deceased population
of NRW (i. e. official death certificate information is available for all
deceased individuals and not only for registered cancer cases). The cancer
registries may use additional death information from periodic linkages with
population registers (only date of death) and from hospitals.
For the mortality evaluation of the German MSP, the epidemiological cancer
registries of NRW, Lower Saxony, and Bavaria, agreed to cooperate and
provide official death certificate information for the linkage with health
insurance claims data.
The reporting of this paper followed the RECORD statement [18 ].
Study Population and Linkage Procedure
In step 1 , based on pseudonymized BARMER data, women aged 40 to 90 years
were selected whose insurance was terminated between 2006 and 2018 and whose
last known place of residence was NRW, Lower Saxony, or Bavaria. Besides women
whose insurance was terminated due to death (TD), women whose insurance was
terminated due to other reasons (TOR) were also considered to evaluate the
differentiation between both reasons of insurance termination in the BARMER
data. We assigned each selected woman a communication ID for the linkage
procedure. Potential cancer diagnoses were not considered in the selection
process, and information on whether women were registered in a cancer registry
was not available at this stage. In step 2 , the BARMER – the only party
with access to unencrypted identifying information of insured individuals –
re-identified the women selected in step 1 and added encrypted person
identifiers according to cancer registry standards. The encrypted information
included, i. a., information on name, address, and date of birth. If more than
one address was available per woman, the BARMER generated multiple records to
potentially increase linkage success. In step 3 , cancer registries linked
the encrypted data to their databases and deleted the encrypted identifiers. In
step 4, we linked the death information to anonymized claims data via
the communication ID. For further information on the approval, data access, and
linkage process, see Langner et al. [5 ].
Statistical analysis
For the analyses, the linkage results were reduced to one record per ID (i. e.
duplicates resulting from the linkage procedure were deleted). We calculated for
both reasons of insurance termination (i. e. TD or TOR) i) the proportion of
women linked to the official death certificate database in NRW, including all
deceased individuals, among all women whose insurance was terminated, and ii)
the proportions of women linked to the official death certificate databases of
registered cancer cases in NRW, Lower Saxony, and Bavaria among all women whose
insurance was terminated. These proportions were compared by region (NRW, Lower
Saxony, and Bavaria), year (2006 to 2018), and age (40–49, 50–59, 60–69, 70–79,
and 80–90 years). In a sensitivity analysis, we restricted the linkage sample to
women with a cancer diagnosis (ICD-10 C00-C97) in the claims data up to three
years before their insurance termination due to death between 2008 and 2018.
Then we calculated the proportions of women linked to the databases of
registered cancer cases within this subgroup. For this analysis, hospital
discharge and ancillary hospital diagnoses as well as outpatient diagnoses coded
as “certain” or “status post” were considered. This analysis was conducted
because it can be assumed that with respect to the linkages with the databases
including only registered cancer cases, the proportions of successful linkages
also depend on the number of cancer cases in the claims data.
Additionally, the agreement between the insurance termination date and the linked
official death date was calculated. As Lower Saxony and Bavaria reported only
the month and year of death, official death dates in all three federal states
were set to the 15th of the reported month. The agreement was
measured in two categories: i) difference of≤31 days, and ii) official death
date>31 days before insurance termination. Official death date>31 days
after insurance termination were ignored/counted as no-matches because we expect
that these cases deceased after insurance termination.
Results
For the linkage procedure, 150,369 (TD) and 96,007 (TOR) women in NRW, 47,472 (TD)
and 42,007 (TOR) in Lower Saxony, and 65,893 (TD) and 49,566 (TOR) in Bavaria were
selected ([Fig. 1 ]). The median age
of the women ranged from 81 to 82 years (TD) and 49 to 51 years (TOR) in the three
federal states, and hardly varied between 2006 and 2018. The proportions of women
with a cancer diagnosis in up to three years before their insurance termination
ranged from 45.24% to 45.92% (TD) and 8.39% to 8.93% (TOR), and increased by less
than five percentage points in all three federal states between 2008 and 2018.
Regarding women with TD, the proportions with cancer increased from the first (40–49
years) to the third (60–69 years) age group, and then decreased to the last age
group (80–90 years), whereas regarding women with TOR, the proportions with cancer
increased from the first to the last age group in all federal states. The BARMER
added encrypted person identifiers for 150,379 (TD) and 96,034 (TOR) records in NRW,
47,491 (TD) and 42,071 (TOR) in Lower Saxony, and 66,490 (TD) and 49,962 (TOR) in
Bavaria. In NRW, official death certificate information was provided for 135,291
(TD) and 303 (TOR) records from the database including all deceased women, and
53,958 (TD) and 118 (TOR) records from the database including only registered cancer
cases. The latter numbers were 18,337 (TD) and 60 (TOR) in Lower Saxony, and 13,330
(TD) and 41 (TOR) in Bavaria. After 16 (TD) and 0 (TOR) duplicates were deleted,
official death certificate information was available for 135,275 (TD) and 303 (TOR)
women in NRW among all deceased women. Concerning registered cancer cases, 0 (TD)
and 0 (TOR) duplicates in NRW, 11 (TD) and 0 (TOR) in Lower Saxony, and 89 (TD) and
0 (TOR) duplicates in Bavaria were deleted. This resulted in 53,958 (TD) and 118
(TOR) women in NRW, 18,326 (TD) and 60 (TOR) women in Lower Saxony, and 13,241 (TD)
and 41 (TOR) women in Bavaria with official death certificate information obtained
from the databases including only registered cancer cases.
Fig. 1 Flow chart with proportions of matches for three federal states
and by reason for insurance termination. 1 Denominator for
displayed percentages in that column; 2 Duplicates are due to
different addresses in the BARMER database; TD: Termination due to Death;
TOR: Termination due to other reasons.
Proportions of successful linkages by region
In NRW, the proportion of women with TD linked to the database including all
deceased individuals, was 89.96% ([Fig.
1 ]), of which 99.92% had a difference of≤31 days between the insurance
termination date and the linked official death date. Regarding the databases
including only registered cancer cases, linked proportions were 35.88% in NRW
(99.93% with≤31 days difference), 38.60% in Lower Saxony (99.97% with≤31 days
difference), and 20.09% in Bavaria (99.80% with≤31 days difference). In the
sensitivity analysis, linked proportions were 75.04%, 78.71%, and 42.27% in the
three federal states.
The proportion of women with TOR linked to the database including all deceased
women, was 0.32% in NRW (94.72% with≤31 days difference). With respect to the
databases including only registered cancer cases, the proportions were 0.12% in
NRW (93.22% with≤31 days difference), 0.14% in Lower Saxony (86.67% with≤31 days
difference), and 0.08% in Bavaria (78.05% with≤ 31 days difference).
Proportions of successful linkages by year
In NRW, the proportion of women with TD linked to the database including all
deceased individuals, increased from 83.14% in 2006 to 93.70% in 2018 ([Fig. 2 ]). Regarding the databases
including only registered cancer cases, linked proportions increased from 28.19%
to 42.29% in NRW, 34.43% to 44.32% in Lower Saxony, and 18.65% to 20.16% in
Bavaria. In the sensitivity analysis, linked proportions increased from 66.38%
in 2008 to 80.74% in 2018, 75.51% to 84.71%, and 39.29% to 40.95%,
respectively.
Fig. 2 Proportions of matches for the linkage between BARMER
health insurance claims data and official death certificate information
by year of insurance termination. RCC: Registered Cancer Cases; NRW:
North Rhine-Westphalia; BA: Bavaria; LS: Lower Saxony, TD: Termination
due to Death.
The proportion of women with TOR linked to the database including all deceased
women decreased from 0.45% to 0.26% in NRW. With respect to the databases
including only registered cancer cases, the proportions decreased from 0.19% to
0.12% in NRW and from 0.55% to 0.01% in Lower Saxony, and 0.25% to 0.04% in
Bavaria.
Proportions of successful linkages by age
In NRW, the proportion of women with TD linked to the database including all
deceased individuals decreased from 95.39% (40–49 years) to 87.70% (80–90
years), whereby the decrease was particularly observed among women≥70 years
([Fig. 3 ]). Regarding the
databases including only registered cancer cases, linked proportions increased
from the first (40–49 years) to the third (60–69 years) age group, where they
were 54.72% (NRW), 56.10% (Lower Saxony), and 32,20% (Bavaria). Subsequently,
the proportions decreased to 26.04%, 29.73%, and 14.35% in the last age group
(80–90 years). In the sensitivity analysis, linked proportions were 87.43%,
88.74%, and 51.81% in the third age group, and 64.07%, 70.14%, and 35.08% in the
last age group.
Fig. 3 Proportions of matches for the linkage between BARMER
health insurance claims data and official death certificate information
by age at insurance termination. RCC: Registered Cancer Cases; NRW:
North Rhine-Westphalia; BA: Bavaria; LS: Lower Saxony, TD: Termination
due to Death.
The proportion of women with TOR linked to the database including all deceased
women increased from 0.08% (40–49 years) to 6.91% (80–90 years) in NRW. With
respect to registered cancer cases, these proportions increased from 0.02% to
2.34% in NRW, 0.04% to 1.44% in Lower Saxony, and 0.01% to 1.28% in Bavaria.
Discussion
This work determined the proportions of successful cause-of-death linkages between
the second-largest German statutory health insurance fund and three cancer
registries. We found that 90.0% of women aged 40–90 years whose insurance was
terminated due to death were linked to the database of all deceased individuals in
NRW. Regarding the databases including only registered cancer cases, linked
proportions were 35.9% in NRW, 38.6% in Lower Saxony, and 20.1% in Bavaria. Linked
proportions increased from 2006 to 2018 and were highest in age group 60–69
years.
Concerning the linkage with the database including all deceased individuals in NRW,
our linked proportion of 90.0% is slightly lower compared to the linked proportion
of 94.7% reached in a previous work [12 ]. This might be explained by the fact that the proportion of linkages
is lower among older compared to younger age groups, and our work included more
older women than the existing study (40–90 years vs. 25–80 years) [12 ]. Furthermore, it should be noted
that the proportion of linkages in our work increased over time due to data quality
improvements and was 93.7% in 2018. The remaining proportion of missing linkages
might be due to deviations in the person identifiers available at the different data
sources [19 ]. Furthermore, insurance
funds have only the mailing address of insured individuals, which is not necessarily
the official registered address documented in the official death certificate. In
some cases, it might also be possible that cancer registries have no information on
deaths. With respect to the agreement between the insurance termination date and the
linked official death date, we observed a difference of≤31 days in 99.9% of the
linked cases, which is nearly the same as the 99.8% reported in the previous work
[12 ].
Regarding the linkages with the databases including only registered cancer cases,
we
observed higher proportions of linked individuals in NRW (35.9%) and Lower Saxony
(38.6%) compared to Bavaria (20.1%). This might partly be explained by a lower
cancer incidence and cancer mortality among women in Bavaria compared to those in
NRW and Lower Saxony (raw incidence and mortality per 100,000 women according to
official data [14 ]
[20 ] in 2013: 527.0 and 221.6 in
Bavaria, 622.9 and 267.6 in NRW, and 611.7 and 249.1 in Lower Saxony). However, it
should also be considered that in our study population the proportions of women with
a cancer diagnosis in up to three years before their insurance termination were
almost identical in the three federal states. Furthermore, our sensitivity analysis
restricted to women with a cancer diagnosis before their insurance termination also
showed higher proportions of linked individuals in NRW (75.0%) and Lower Saxony
(78.7%) compared to Bavaria (42.3%), which suggests that the observed differences
must also be due to other aspects. A further explanation are differences between
cancer registries concerning inclusion of cases with non-melanoma skin cancer
(ICD-10 C44). While these cases are registered systematically in NRW and Lower
Saxony, registration in Bavaria is only partial, resulting in a smaller database.
Another reason might be different degrees of data quality improvements in the three
federal states between 2006 and 2018: While the linked proportion increased by 14.1
(sensitivity analysis 14.4) percentage points in NRW and 9.9 (9.2) percentage points
in Lower Saxony, it increased by only 1.5 (1.6) percentage points in Bavaria.
Furthermore, due to legal restrictions the cancer registry in Bavaria was unable to
perform a linkage between cancer cases and the population register between 2009 and
2016, which resulted in less accurate data on changes of names, places of residence,
and deaths. Regarding our main analysis, it must generally be considered that we
calculated the proportion of successful linkages among the total population of women
with insurance termination due to death because separation of individuals registered
in a cancer registry is not possible in claims data. Given that i) according to data
of the cancer registry of NRW only 53% of all deaths between 2006 and 2018 among
women aged 40 to 90 years were registered cancer cases and ii) deviations in person
identifiers and address data, as well as missing death information among registered
cancer cases exist, the observed proportions of successful linkages in the linkage
with the databases including registered cancer cases were below 50% in all three
federal states as expected. However, the proportions of successful linkages
decreased from the third to the last age group in all three federal states in both
the main and sensitivity analysis, which indicates that deviations in address data
might be particularly common among women aged≥70 years. Especially in cases of
end-of-life care in institutional settings, deviations between the mailing address
of insured individuals and the official registered address documented in the
official death certificate are to be expected.
With respect to women whose insurance was terminated due to reasons other than death,
the proportions of linked individuals were below 0.4% regarding the database of all
deceased individuals in NRW and below 0.2% with respect to the databases including
only registered cancer cases in the three federal states. Because the expected
proportion was nearly 0%, this indicates a valid claims-based differentiation
between insurance terminations due to death and other reasons in the BARMER data.
Nevertheless, it must be considered that in claims data in some cases information
on
deaths is missing in the core data or is only indicated in hospital data as reason
for discharge [21 ]. In our study, the
latter was the case for 33.3% of the 303 linked individuals from NRW whose insurance
was terminated due to reasons other than death.
Overall, our results show that claims data can be enriched with official death
certificate information from cancer registries. The data can be used for studies
addressing cancer mortality, such as the evaluation of the German MSP. When
conducting such studies, it should be borne in mind that causes of death cannot be
linked for all deceased women, and the proportions of linkages differ by region,
year, and age. Therefore, claims-based algorithms should also be considered to
enrich claims data with causes of death [6 ]
[22 ].
Strengths and limitations
A major strength of this work is that it provides detailed information on the
proportions of successful cause-of-death linkages between the second-largest
German health insurance fund and three federal cancer registries for a period
covering 13 years. A further strength is that more than 260,000 women whose
insurance was terminated due to death as well as nearly 188,000 women whose
insurance was terminated due to other reasons were included in the linkage
procedure.
There are, however, some important limitations. First, information on all
deceased women was available in only one federal state (NRW). This state has,
however, the largest population of all German federal states [14 ]. Second, we used the month and
year of death to calculate death dates but were unable to consider the day of
death. However, Langner et al. showed that in 99.8% of linked cases, the date of
insurance termination and official death was identical (96.5%) or differed by≤31
days (3.5%) [12 ]. Third, because
claims data contain no information on cancer registration, we calculated the
proportion of successful linkages among the total population of women with
insurance termination due to death. Therefore, the expected proportions of
successful linkages with the databases including registered cancer cases were
below 50%. Fourth, our study includes only one of the currently 96 German
statutory health insurance funds [23 ] and ignores privately insured individuals, who represent 10% of
the German population [24 ].
Structural differences between insurance funds and statutorily and privately
insured individuals may limit the generalizability of our results [25 ]
[26 ]. The considered health
insurance fund is, however, large and insures a sizeable proportion of women.
Fifth, we conducted linkages with only three of the 15 German cancer registries.
The included registries are, however, from comparatively large federal states
representing 46.7% of all women aged 40–90 years. Finally, we conducted
probabilistic linkages based on pseudonymized identification variables (control
numbers) because a unique person identifier was not available in all data
sources. In the future, however, cause-of-death linkages between German
insurance funds and cancer registries may also be possible based on health
insurance numbers, which should result in higher proportions of successful
linkages.
This work showed that claims data of the second-largest German statutory
health insurance fund can be enriched with official death certificate
information from cancer registries. The linked data will be used for the
mortality evaluation of the German MSP. Because causes of death could
not be linked for all deceased women and the proportions of linkages
differed by region, increased over time, and was highest in age group
60–69 years, claims-based algorithms will also be used to identify
causes of death.
Fundref Information
Bundesamt für Strahlenschutz mit Mitteln des
Bundesministeriums für Umwelt, Naturschutz, nukleare Sicherheit und
Verbraucherschutz (BMUV), des Bundesministeriums für Gesundheit (BMG) und der
Kooperationsgemeinschaft Mammographie (KoopG; getragen durch den Spitzenverband der
gesetzlichen Krankenkassen und die Kassenärztliche Bundesvereinigung) — 3617S42402
3621S42410.
