Patients at high risk for a severe clinical course of COVID-19 — small-area data in support of vaccination and other population-based interventions in Germany

The study was based on nationwide pseudonymized outpatient claims data encompassing all German statutory health insurances from the years 2010 to 2019 collected in accordance with §295 of the Code of Social Law V (Sozialgesetzbuch, SGB V) [13]. This database is a comprehensive collection of administrative outpatient data of all SHI individuals (87% of the entire German population). Besides sociodemographic characteristics such as age, sex and place of residence, the data include, amongst others, information on billed medical services and diagnoses as well as physician-related characteristics such as specialist group and practice location. Diagnoses are coded according to the International Statistical Classification of Diseases and Related Health Problems, 10th revision, German modification (ICD-10-GM). In addition, diagnoses from German outpatient care include a modifier describing the diagnostic certainty (‘assured’, ‘suspected’, ‘status post’, ‘excluded’) [14]. The study population included individuals aged ≥15 years (N = 61,533,884).

The selection of pre-existing conditions was based on previous findings on risk factors for severe course of COVID-19 disease related to chronic health conditions reported in international studies. We included chronic conditions that were associated with a markedly increased age-adjusted risk in several epidemiological studies of COVID-19 patient groups or the general population based on selective literature search. Underlying chronic conditions for which there was no clear evidence in the selected literature were not considered. A markedly increased risk was generally assumed, if the risk of the outcomes hospitalization, intensive care, and/or death was increased by the factor 2 or more in age-adjusted analysis in at least one of the identified studies. The prognostically relevant chronic conditions included for the definition of high risk groups are given in Table 1.

In the population of all SHI individuals in the year 2019, occurrence of the included chronic conditions was assessed on the individual level. For the majority of diseases or disease groups this was done by using the so called M2Q-criterion. Accordingly, patients were defined as having prevalent chronic diseases, if they had a diagnosis coded with the diagnostic certainty ‘assured’ for the respective condition or disease group in at least two quarters of the year 2019. For the identification of cases per disease entity according to the M2Q-criterion, it was irrelevant if identical or different diagnoses from the list of ICD-codes per disease group were assigned in at least two quarters. For disease groups with assigned subgroups (cardiovascular diseases and other neurological diseases) the M2Q-criterion had to be met for at least one of the respective subgroups.

For solid tumors, hematological tumors, stroke or post-stroke conditions, cerebrovascular precursors as well as transplantations and post-transplantation conditions the case definition was modified. For the assessment of stroke or post-stroke conditions and cerebrovascular precursors according to the M2Q-criterion, diagnoses coded with the diagnostic certainty ‘status post’ were considered in addition to ‘assured’ diagnoses. The same approach was applied for transplantations and post-transplantation conditions. In contrast, solid tumors were included as incident diseases, if the M2Q-criterion applied for at least one of the considered ICD-three-character-codes in 2018 and/or 2019 and if there was no diagnosis coded with diagnostic certainty ‘assured’ of a solid tumor in the years 2010 to 2017. Current evidence suggests an especially increased risk associated with hematological tumors [23], even with a first occurrence of the disease in the past [5]. Hence, patients who met the M2Q-criterion at least once for one of the considered ICD-three-character-codes in the years 2015 to 2019 were also included.

The continuously growing study base indicates the high relevance of age and specific chronic conditions for the prognosis of severe COVID-19 courses. Early after the onset of the pandemic, a higher age emerged as the most important predictor for the need of intensive care due to COVID-19 [27, 28]. Furthermore, epidemiological studies showed a largely consistent pattern of a markedly increased risk associated with the chronic conditions included in this study. However, only limited evidence is available on the interaction of age and chronic conditions as well as chronic conditions with each other. The majority of the reviewed epidemiological studies and meta-analyses did not consider the interaction of age and chronic conditions in the statistical models applied for the risk estimation. Regarding the interplay of several comorbidities, individual studies confirm a clinically plausible accumulation of risks from single chronic conditions [29, 30].

Under the pragmatic assumption of essentially aligned additive associations of age and comorbidity related risk, a high risk of severe COVID-19 courses was postulated for the following groups and classified respectively:

Based on the profile of the included chronic conditions on the individual patient level, we calculated the prevalence of at least two prognostic relevant chronic conditions for the age group 15 to 59 years and of at least one for the age group of 60 to 79 years on district level. As analyses on interacting effects of the risk factors hypertension and age suggested a decreasing risk of hypertension with increasing age in a large British population-based study [5], hypertension was not classified as a chronic condition with high prognostic relevance in the age group 60 to 79 years.

In addition to the assessment of population sizes of the three predefined risk groups, we calculated the total number of individuals with at least one of the relevant chronic conditions in the age group 15 to 59 years. To assess robustness of our results based on morbidity profiles captured from SHI claims and population statistics of 2019, analyses were repeated using 2018 data.

The estimation of the regional populations’ size among German inhabitants with a high risk for a severe COVID-19 course was conducted for the three risk groups. The risk groups 1 and 2 were created based on an extrapolation of the respective population-based prevalence of risk group in the SHI population to the population of all German inhabitants in the respective age group. A pragmatic fundamental assumption was that the regional age-specific morbidity in the SHI population was similar to that of the general population. The risk group 3 included the complete population aged 80 years and older irrespective of the presence of chronic conditions. The extrapolation was conducted using population data on administrative district level from the German regional database of the federal and state statistical offices on www.​regionalstatisti​k.​de.

A specific strength of our research results from the coverage of insured people encompassing all SHI individuals and the supra-regional character of the outpatient claims data used. Due to a high frequency of comorbid occurrence of included relevant chronic conditions for risk classification, the use of results from previous primary data studies with usually a focus on single diseases comes with the risk to overestimate the size of the vulnerable population. At the same time, the assessment of the combined occurrence of single risk factors from the entire spectrum of the relevant chronic conditions allows the description of accumulated risks on the level of individual patients.

Some limitations have to be mentioned. The data set of the present study did not capture chronic conditions associated with a severe COVID-19 course that have been diagnosed in inpatient settings only. However, sole inpatient treatment is assumed to be relatively rare. Furthermore, patterns of distribution of morbidity in the SHI population have been extrapolated to the general German population, which in 2019 also comprised about 11% of inhabitants covered by private insurance. Due to an assumed somewhat lower prevalence of relevant risk factors in the privately insured population, which has been shown for some chronic conditions, our findings may have overestimated the size of the population with high vulnerability to an unknown but at most moderate extent. In the used SHI claims data, pseudonymized patient identifiers are formed by assigning unique integers to every distinct combination of a patient’s first and last name and a patient’s date of birth. A patient’s pseudonym can change permanently in case of name changes (e.g. by marriage). Furthermore, a patient’s pseudonym may temporarily change due to occasionally occurring erroneous data entries in outpatient physicians’ offices. As a result with regard to prevalence estimates both numerator (outpatients aged ≥15 years with specific diagnoses in two quarters of 2019) and denominator (all outpatients aged ≥15 years) likely were subject to double counting of a minority of patients with changes of attributes used to form patients‘pseudonyms during 2019. As both numerator and denominator are simultaneously affected to a similar extent, it can be assumed that prevalence estimates closely reflect the true morbidity in the SHI population.

The current study aimed to give an estimate of the size of populations with a high risk of a severe COVID-19 course to allow appraisal of regionally varying needs for vaccination due to morbidity differences from pre-existing chronic conditions. Due to a lack of detailed clinical data in insurance claims the employed algorithm features limited capabilities to differentiate between subgroups of patients depending on the severity and course of chronic conditions and hence may result in misclassification of individual risks in some patients. In case of solid cancers the population at increased risk was likely underestimated since only patients with incident disease in 2018 or 2019 were captured, but patients with cancer relapse in 2018 or 2019, who were already treated for cancer sometime during the preceding seven year period were excluded from case definition.

Since only limited evidence exists on the interaction of risks of age and chronic conditions as well as chronic conditions with each other, risk groups were classified using a pragmatic approach. Accordingly, essentially aligned additive associations of age and comorbidity related risk were assumed. This approach allowed to give estimates on the size of populations with high vulnerability incorporating cumulation of individual risks. As the nature of interactions of comorbidities is unknown, this approach likely oversimplyfied the interplay of single risk factors, which may be rather mutiplicative than additive in many cases.