We updated influenza-associated excess mortality and hospitalization estimates in Germany for all seasons from 1995–1996 to 2017–2018 using weekly data and investigating potential patterns in age and cause (for mortality only). The publicly available data from the FDZ and the RKI influenza working group appear to be suitable for addressing the limitations in the RKI’s mortality model named by the RKI itself. Our model estimates aim to complement RKI's results, by providing an additional level of detail needed for medical decision-making. We found similar excess death results using weekly instead of monthly death figures. Our conservative model calculated an annual weighted average of approximately 7600 (9.32 per 100,000 population) compared to RKI’s 8500 (10.46 per 100,000; self-calculated average [32]) excess all-cause deaths. On average, more than 95% of estimated excess mortality occurs in those aged ≥ 60 years (7300; 35.25 per 100,000) and, within this group, more than 43% of these excess deaths occurred in patients 80–89 years (3200; 95.40 per 100,000). The estimated annual average excess mortality from 1998–1999 through 2017–2018 across all ages was around 3700 deaths (4.56 per 100,000), 2000 (2.41 per 100,000), and 900 (1.14 per 100,000) with underlying circulatory, respiratory, or P&I cause, respectively.

In general, there are considerable differences in the statistical methods used in various studies to assess influenza-associated mortality and the results vary widely, with estimates increasing with age [10]. The systematic review from Li et al. [10] found values between – 0.3 to 1.3, 0.6–8.3 and 4–119 respiratory deaths per 100,000 population for the different age groups, defined as children, adults, and older adults. Our values (0.18 per 100,000 (0–17 years), 0.70 per 100,000 (18–59 years), and 35.25 per 100,000 (≥ 60 years)) are in line with this and thus confirm the results found there.

We estimated an annual average of around 36,000 (43.90 per 100,000) excess hospitalizations attributed to influenza, compared to RKI’s 17,000 (20.74 per 100,000) hospital admissions. This suggests a potential, yet important, underestimation of the RKI model, due to their methodological approach. Influenza-associated excess hospitalization in Germany is estimated based on data from a surveillance system that monitors medically-attended acute respiratory infections (MAARI) in primary care practices. This involves two steps: first, excess consultations in Germany are estimated using a generalized additive regression model (GAM), with the age-specific weekly MAARI rate as a dependent variable [13, 30], then, second, the proportion of hospital admissions of all MAARI from the surveillance are determined and multiplied by the number of excess consultations, resulting in estimated excess hospitalization [30, 31]. We believe this two-step approach has a major limitation. The surveillance data come from office-based physicians; however, many referrals are not made by the general practitioners themselves, but rather by an on-call emergency physician or in the emergency room of a hospital. Therefore, the use of the proportion of hospital admissions among all MAARI may substantially underestimate influenza-related hospitalization burden as shown here. As in the mortality model, excess hospitalizations affect those aged ≥ 60 years most, with 57.49% of excess hospitalizations in this age group (24,300; 117.26 per 100,000). 32.96%, or around one-third, of all excess hospitalizations occurred in patients aged 18–59 (13,900; 29.74 per 100,000), which also has a major impact on the economy and should not be neglected. In Germany, the RKI estimates that 5.3 million people were unable to work due to influenza in 2017–2018, with the highest proportion in patients aged 35–59 at 2.9 million [40].

Our results clearly show that the main burden of influenza is in the elderly. For years, an influenza vaccination coverage rate of 75% and higher has been called for worldwide [41], and especially in the EU [42], for this group. However, current data show [43] that, in 2022, the rate for those aged ≥ 60 years in Germany was 43.30%, falling far short of the target.

Influenza’s burden varies by season, which is likely related to variations in circulating viruses, vaccine effectiveness, and annual influenza epidemic timing. We found the lowest excess all-cause mortality occurred in 2005–2006 and 2009–2010, and the highest in 2016–2017 and 2017–2018. In contrast, excess all-cause hospitalization rates were lowest in 1998–1999, 2000–2001, and 2007–2008, and highest in 2010–2011 and 2016–2017.

Unlike most other methods, the RMDM does not use regressions, but relies solely on the recurrent relative mortality distribution pattern. It does not require a certain number of observations and does not use a proxy for influenza activity. Compared to other much more complex models, it is comparatively simple, practical, and does not need advanced mathematical expertise. A current internationally applied example of regression models with an influenza activity proxy is the FluMOMO model [16, 17, 37]. It is a time series regression model using a Poisson distribution, with correction for overdispersion, and includes influenza activity and extreme temperature as independent variables. Serfling-type models are also used internationally, to identify the cyclical components in the time series using Fourier terms [20, 21]. As with the RMDM, no proxy for influenza activity is used. All the models mentioned use the same definition of excess values (observed minus expected number) but calculate the baseline differently. Regression models require a minimum sample size (unnecessary for the RMDM), and regression models’ overall complexity is significantly increased and may be difficult for non-statisticians to apply.

Like the RKI model, our model is also subject to some limitations. Since we define excess as the difference between observed and expected values, we attribute every deviation from expected mortality/hospitalization within the influenza season to influenza. Other viruses that regularly circulate in addition to influenza, as well as other factors, might cause us to overestimate the excess in given weeks. An overlap with other virus epidemics, such as respiratory syncytial virus (RSV)—which also has a substantial impact in the winter season and could cause similar health results in some seasons—could not be ruled out. If possible, this should be controlled in a future model. Similarly, environmental temperature can have a substantial impact on mortality/hospitalization and should therefore be incorporated into predictive models to improve accuracy. Furthermore, smaller subgroups and weeks with fewer deaths on average might be more susceptible to random outliers. In very small groups, there is also always the possibility that excess—not revealed by the RKI model—could be random and not influenza-associated.

Analogous to the RKI method, we have set negative values to zero in the calculation of the excess. This a priori assumes that influenza can only increase the rate and does not decrease it. This can lead to a situation where positive random deviations are no longer neutralized by negative random deviations and the model thus produces an excess that does not exist in reality. With the awareness that we assumed a positive association between influenza and excess mortality/hospitalization, we wanted to maintain as many similarities to the RKI model as possible. So far, we have treated influenza by its presence alone (i.e., binary); other models incorporate the type of virus [14] or the strength of an influenza wave [16].

Like any other study based on secondary data, data collection was not adapted to our research questions. The information is limited to billable services and therefore influenced by coding quality of diagnoses and procedure; for example, diagnoses for patients are coded by healthcare providers that do not entirely correspond to the clinical picture of the patient.