In this epidemiological study on the risk of SPLC, we demonstrated a bias introduced by the standard approach of calculating SIR for same-site SPC in cancer registries that use IARC/IACR MP rules. Our simulations showed an underestimation of true SIR by approximately 30% in both sexes because general population reference rates include tumors that cannot be observed in the at-risk population.

To minimize this bias, we proposed the method of histology-specific SIR and presented adjusted numbers to estimate the relative risk of SPLC in LC survivors in Germany. Using this method, the risk of being diagnosed with an SPLC was 2.98-fold for female LC survivors and 1.15-fold for male survivors compared to the general population. These numbers are substantially higher than previously published unadjusted SIR estimates using the standard approach (i.e., 2.06 and 0.83 for female and male LC survivors, respectively) [7].

Our validation analysis showed that the risk estimates for SPLC under SEER MP rules and the risk under IARC/IACR MP rules converge when using the proposed histology-specific compared to the standard approach. Moreover, the SPLC risk in Germany and the USA became comparable under the new method. In the descriptive analysis, we observed a similar incidence of primary LC, but an eight times higher crude incidence of SPLC in the USA compared to Germany, due to different registration rules and population structure. We narrowed this gap to an about twofold increased risk in the most comparable setting of histology-specific SIR counting international primaries only. Thus, we showed that the proposed histology-specific method makes the SIR metric more robust against differing or unclear MP rules and, therefore, more suitable for cross-country comparisons.

With the availability of long-term cancer registry data from SEER, relative risk estimates and survival analyses were predominantly published for the USA. Thakur et al. used 13 SEER registries for the period 1992 to 2007 and found a SIR estimate of 4.85 (4.66–5.05) in women and 3.38 (3.24–3.52) in men [9]. Our SIR1raw estimates for a later period with more included regions were slightly higher for both sexes but—as expected for the same data source—showed very similar risk increases for younger age and adenocarcinoma index LC.

Other national population-based estimates of SPLC risk among LC survivors are sparse but include a study by Barclay et al. from the England cancer registry that showed an overall SIR for SPLC of 1.47 (not stratified by sex) for the same period, also using IARC/IACR MP rules and very similar case inclusion as our study [8]. They included almost 30% of unspecified LC diagnoses, which might explain their low overall unadjusted SIR estimate.

For Germany, we observed that using histology-specific calculation of risk for SPLC resulted in very different conclusions than the standard method. The previously published risk reduction for male LC survivors [7] turns into a risk increase in our updated analysis. Similarly, for the most frequent subtypes of LC, the risk of SPLC increases from a protective or null effect to a significant risk increase for LC survivors compared to the general population. In line with previous findings from Germany, the USA, and England, females have a higher relative risk for SPLC than males [7,8,9]. The results of bias simulation apply to other PBCRs using IARC/IACR MP rules because they depend only on the distribution of histology groups, which are similar across countries [29].

The SIR is an established measure in cancer epidemiology and beyond to identify at-risk populations and test etiologic hypotheses [19]. However, our analysis revealed the problem that the SIR metric can be seriously biased when observed and expected counts are based on different logics. Other examples are handling the intensified screening period of cancer survivors shortly after diagnosis [19], including DCO cases in the expected count [30], or stratifying by race.

A critical discussion in the analysis of SPC risk is the question of what constitutes biologically independent cancers. Although there might be a pathological answer, strict (IARC) and less stringent (SEER) rules for registering multiple primary cancers indicate different approaches. While IARC/IACR MP rules were historically focused on excluding potential metastases or recurrences, they give less importance to shared risk factors such as smoking, which can lead to preventable new cancers with the same appearance as the previous disease. Therefore, the SEER approach of registering SPC independent of histological differences and instead taking time into account has its merits. From the patient’s perspective, it is less relevant if, after a phase of complete remission, a new cancer of the same subtype occurs in the same site or whether this is an unexpected late recurrence. Both necessitate acute treatment according to the most recent guidelines and come with substantial psychological stress. From an epidemiological perspective, the differentiation between what constitutes a true SPC and late recurrence can be seen as arbitrary. Both are relevant to evaluating the disease risk of cancer survivors and play a role in the discussion about the need for increased surveillance. In particular, contralateral lung cancers are often underreported and instead seen as the spread of the original cancer [31]. On the other hand, it has been shown that certain multifocal tumors in some organs are likely to originate from the same transformed cells [32] and thus would not have been preventable. A potential solution to this dilemma could be to enrich existing PBCR with more information on metastases and recurrence and report the burden of both. Furthermore, improving the quality of cancer registries, e.g., increasing the share of microscopically verified cases, would reduce the role of unspecified LC and thus increase the precision of SPC differentiation under IARC/IACR MP rules.

This analysis also highlights the need to improve the clarity of IARC/IACR MP rules that have not been updated since 2004 and do not reflect improvements in (more fine-grained) histological classification since then. A simplified version using the broad LC categories recently published in the WHO Reporting System for Lung Cytopathology could be the common basis for a classification update [20, 33].

The importance of methodological choices on the validity of the SIR metric to determine and compare group-specific cancer risk has already been discussed. Crocetti et al. highlighted that we need to be clear about which cases to include in the observed count and the importance of applying the same logic when making cross-country comparisons and comparing extensively screened patients to the general population reference rates [34]. Our research extends this concern and highlights that including expected cases in the denominator of the SIR metric while these cases can never be observed—due to MP rules forbidding their registration—will introduce bias and result in an underestimation of cancer risk.

This study is the first nationwide analysis to provide reliable estimates of SPLC risk in German LC survivors. It adds to the sparse international numbers, as PBCR-based estimates have only been published for the USA and England. We newly developed the method of histological subtype-specific SIR that can reduce bias caused by using general population reference rates in determining the relative risk of same-site SPC for cancer registry data based on IARC/IACR MP rules. The method increases the comparability of SIR across countries when different MP rules are applied, information on registration practices is missing, or registration practices vary over time within a single PBCR. Furthermore, our study is the first to compare the relative risk of SPLC between Germany and the USA and provide estimates that allow SEER data to be used in international comparison.

Our proposed method has several limitations. In the situation in which PBCR data is incomplete, i.e., incident SPCs are not recorded in the registry, the proposed method still underestimates the SIR. This is especially relevant for IARC-compliant registries where SPCs are not recorded if the histology of the index tumor is unspecified. Moreover, the method works best for cancer entities that show a uniform distribution of incidence across histological subtypes. For example, the subtype is not informative in breast cancer, where adenocarcinoma comprises more than 90% of all malignancies. We encourage researchers with access to cancer registry data from other countries to test the proposed method. With regard to the estimation of SPLC risk for Germany, our study results are limited by varying completeness and quality of regional PBCR data—in particular, the existence of tumors without histological confirmation, a high share of DCO diagnoses, and incomplete follow-up of patients in cases of migration or partial record linkage with death registry data. Finally, our validation analysis also showed that for registry data under SEER MP rules, the histology-specific SIR generally suggests a lower risk than the unadjusted SIR (except primary SCLC). This could be an underestimation of risk if the true SIR of same-histology SPLC is higher than for different-histology tumors.