Emerging trends of therapy related myeloid neoplasms following modern cancer therapeutics in the United States

Clonal hematopoiesis (CH) is a risk factor for the development of therapy-related myelodysplastic syndromes (tMDS) and acute myeloid leukemia (tAML). Adoption of targeted-immunotherapeutics since 2011, may alter the risk of CH progression to tMDS/AML. To study this, we evaluated risk of tMDS and tAML in 667 588 ≥ 1-year survivors of non-small cell lung cancer (NSCLC), renal cell carcinoma (RCC), melanoma and multiple-myeloma (MM) diagnosed during: 2000–2005, 2006–2010 and 2011–2016. The risk of tMDS increased significantly after NSCLC across all time periods (Ptrend = 0.002) while tAML risk decreased from 2006–2010 to 2011–2016, coinciding with increasing use of non-chemotherapeutic agents. tAML risk after RCC decreased (Ptrend = 0.007) whereas tMDS risk did not significantly change over time. After melanoma, tMDS and tAML risks were similar to the general population. tMDS and tAML risk after MM increased from the first to second time-period, however, only risk of tMDS decreased during last period. We report diverging trends in the risk of tAML and tMDS after adoption of modern cancer therapies for specific cancers. It is imperative to further explore impact of contemporary treatment strategies on clonal evolution. Modern treatments via their discrete mechanism of actions on pre-existing CH may alter the risk of subsequent tMDS and tAML.

Statistical analyses of tMDS and tAML. tMDS and tAML risks were calculated using standardized incidence ratios (SIRs) and 95% confidence intervals (CIs) in SEER*Stat (version 8.3.6). Exact Poisson methods were used to calculate P-values for trends and heterogeneity. SIRs were calculated by dividing the observed number of MDS and AML cases by the number of expected MDS and AML cases in the general population. The expected number of MDS, AML cases were calculated by multiplying the stratified general population incidence rates [by 5-year age groups, race, sex, and calendar year of diagnosis (2000-2005, 2006-2010, 2011-2017)] by the person-time at risk of the cohort 19 . Subjects were followed to earliest of: tMDS or tAML diagnosis, last followup, age 85 years, death, 5 years, or end of study (December 31, 2017). In order to reduce surveillance bias, we excluded tMDS and tAML cases that occurred in the first year of follow-up due to increased surveillance and those patients over the age of 85 years due to decreased surveillance. Additionally, we conducted a sensitivity analysis truncating follow-up on December 31, 2012 allowing for 5 years of follow-up to reduce survival bias in the most recent cohort.

Multivariable Poisson models by first primary cancer.
To assess for SIR trends and heterogeneity by age, sex, race, chemotherapy, radiation, and calendar time period, we conducted multivariable Poisson models by first primary cancer (Epicure version 2.00.02, Risk Sciences International 20 . Specifically, P trends and P heterogeneity values for Tables 2, 3, Supplemental Tables 3, 4 and P trends in Figs. 1, 2, 3 and 4 were computed using these methods. Models were adjusted for age, sex, and race. The log of the expected number of cases was included as an offset to further indirectly adjust for attained age and calendar year 21 . Two-sided P values to test for heterogeneity and trend were computed using a likelihood ratio test derived from Poisson regression models comparing Figure 1. Risk of tMDS and tAML after NSCLC by stage and time period of NSCLC diagnosis. CI confidence interval, NSCLC non-small cell lung carcinoma, SIR standardized incidence ratio, tMDS treatment-related myelodysplastic syndrome, tAML treatment-related acute myeloid leukemia. 'tAML' and 'tMDS' inside the figure represents number of patients diagnosed with t-AML or tMDS, in the respective periods.   Risk of tMDS and tAML after RCC by time period of RCC diagnosis. CI confidence interval, RCC renal cell carcinoma, SIR standardized incidence ratio, tMDS treatment-related myelodysplastic syndrome, tAML treatment-related acute myeloid leukemia. 'tAML' and 'tMDS' inside the figure represents number of patients diagnosed with t-AML or tMDS, in the respective periods.    Fig. 1). Risk for tAML after a diagnosis of localized and distant NSCLC increased across the first two time periods and then remained the same (localized: P trend = 0.17; distant: P trend > 0.2). After conducting sensitivity analyses limiting patients' diagnoses to on or before December 31, 2012 to limit bias in the most recent cohort, risk of tAML after distant NSCLC decreased in the most recent cohort (SIR = 3.02, 95% CI = 1.11-6.58), similar to tAML risk after regional stage NSCLC patients (Supplemental Table 3).
The risk for development of tMDS was elevated after a diagnosis of NSCLC (SIR = 2.16, 95% CI = 1.86-2.49; Table 3). This risk remained elevated after all stages of NSCLC (localized: SIR = 1.95, regional: SIR = 1.  Table 2, Fig. 2). Similarly, there were no trends in the risk for tMDS development after a melanoma diagnosis across and within these time periods (SIRs = 0.80, 0.85, 1.11; Table 3, Fig. 2; P trend = 0.2). Table 2. Standardized incidence ratios for tAML by age, sex, race, and initial diagnosis year among ≥ 1-year adult first primary NSCLC, melanoma, RCC and MM survivors in 17 SEER registries, 2000-2017. MM multiple myeloma, NSCLC non-small cell lung carcinoma, O observed, RCC renal cell carcinoma, SEER Surveillance, Epidemiology and End Results Program, SIR standardized incidence ratio, tAML treatmentrelated acute myeloid leukemia, 95% CI 95% confidence interval. *P < 0.05. P values to test differences in the SIRs were computed using a likelihood ratio test derived from Poisson regression models stratified by age at first primary neoplasm, sex, race, initial diagnosis year, and stage of NSCLC. Categories with < 5 observations were not specified to maintain patient confidentiality. P-values do not include categories with < 5.  Fig. 3, Table 2). In contrast, tMDS risk after RCC demonstrated a non-significant increase over time (SIRs = 0.78, 1.20, 1.38, P = 0.12, Table 3, Fig. 3).
Multiple myeloma. tAML risk after first primary MM did not significantly change across time periods (P trend > 0.2; Table 2, Fig. 4), despite the change from chemo-based regimens to immunomodulatory regimens. When the initial MM diagnosis year was limited to 2012 to allow for a full 5 years of follow-up, tAML risk in the most recent time period increased considerably (SIR = 7.26, 95% CI = 4.49-11.09) although the trend remained non-significant (P trend = 0.18; sTable 3). tMDS risk also did not significantly change over time (SIRs = 3.61, 4.98, 3.53; P trend > 0.2; Table 3, Fig. 4), despite the change from chemo-based regimens to immunomodulatory regimens. When the initial MM diagnosis year was limited to 2012 to allow for a full 5 years of follow-up, tMDS risk was lower in the most recent time period than previous years (SIR = 2.27, 95% CI = 0.98-4.47; sTable 4), in contrast to tAML risk after MM. Table 3. Standardized incidence ratios for tMDS by age, sex, race, and initial diagnosis year among ≥ 1-year adult first primary NSCLC, melanoma, RCC and MM survivors in 17 SEER registries, 2000-2017. MM multiple myeloma, NSCLC non-small cell lung carcinoma, O observed, RCC renal cell carcinoma, SEER Surveillance, Epidemiology and End Results Program, SIR standardized incidence ratio, tMDS treatmentrelated myelodysplastic syndrome, 95% CI 95% confidence interval. *P < 0.05. P values to test differences in the SIRs were computed using a likelihood ratio test derived from Poisson regression models stratified by age at first primary neoplasm, sex, race, initial diagnosis year, and stage of NSCLC. Categories with < 5 observations were not specified to maintain patient confidentiality. P-values do not include categories with < 5. Based on a total of 810 cases, tMDS/tAML occurred more often than expected in NSCLC (SIR 2.38; 2.14-2.63) and MM (SIR 4.46; 3.81, 5.17). The combined risk increased significantly after NSCLC across the three time periods (P trend < 0.001) but not after MM (> 0.20). After initial chemotherapy for NSCLC and MM, combined tMDS/AML risk was elevated (< 0.001). For both combined melanoma and RCC, despite adequate number of observed tMDS/tAML cases (157 and 118 respectively), no particular trends were noted (SIRs similar to general population), speaking to the non-conventional cancer therapy approaches used in the treatment of these cancers.

Multivariate Poisson regression models to evaluate observed differences in SIRs.
In analyses by age, SIRs for tMDS and tAML were highest among patients diagnosed with first primary NSCLC and MM at younger ages (P trend ≤ 0.01 for all; Tables 2 and 3). In analyses by race, risk for tAML after NSCLC and tMDS after RCC was highest among those of black race compared to white/unknown and other races (tAML after NSCLC: P heterogeneity = 0.02; Table 2, tMDS after RCC: P heterogeneity = 0.01; Table 3). No differences in risk existed for tAML or tMDS by sex. Chemotherapy and/or radiotherapy exposure significantly heightened risks for tMDS after NSCLC as well as tAML after NSCLC compared with no therapy. The risk for tAML after MM in patients initially treated with chemotherapy was significantly elevated (P heterogeneity < 0.001); however, risk for tMDS did not differ by initial chemotherapy treatment (P heterogeneity > 0.2). There were too few patients who received chemotherapy or radiation as a part of their initial therapy to evaluate risk of tAML or tMDS after RCC or melanoma.

Discussion
With substantial shifts in treatments over the last two decades, we have demonstrated for the first time changes in tMDS and tAML risk over time that highlights distinctions between tMDS and tAML trends. Specifically, tAML risk decreased after NSCLC, and RCC in the most recent time period and increased after myeloma. Additionally, an increase in tMDS risk after NSCLC, as well as RCC and decrease after MM over time was noted. The following factors could contribute to these trends: (i) Decreasing utilization of leukemogenic agents that are specifically associated with tAML (ii) underlying molecular uniqueness of tMDS and tAML which may be differentially sensitive to various targeted and immune agents (iii) host specific factors such as germline predisposition, smoking etc., unaccounted in this study but randomly controlled by comparison to general population (iv) relative survival changes over time. Additionally, improved case ascertainment of MDS diagnosis in SEER registry over time would not explain these trends, as it would lead to consistent increase in tMDS after all tumor types over progressive time periods. Similar to prior studies 5,6 , we noted that combined tMDS/tAML risk continues to be high after specific cancers, NSCLC and MM. This effect is likely driven by increased tMDS risk after NSCLC and tAML risk after MM, as discussed below. The concordance of the timeline of novel therapy adoption (sTable 1) and changing trends of tMDS and tAML is intriguing and perhaps reflects impact of these agents on myeloid clones that preexist before disease presentation. Mutational profile of tAML has overlap with AML arising from antecedent myeloid neoplasia 23 . While there is a certain overlap of molecular make up between these two entities, key distinctions between tAML and tMDS remain to be explored; an active area for continued research. Mutations of RAS pathway genes (PTPN11, NRAS, KRAS, FLT3) are predominant in tAML in comparison to tMDS wherein mutations such as TET2, ASXL1, SRSF2 and SF3B1 are common 17,18 . Mutations in these genes (pre-tAML clones) have been reported to predate overt myeloid neoplasms with increasing variant allele frequencies (VAFs) over time, from baseline to diagnosis 4,24,25 . Several widely applied targeted agents (sunitinib, sorafenib, everolimus, vemurafenib, erlotinib etc.) act via RAS pathways 26 and could potentially influence clonal propagation of these pre-malignant tAML clones. Since, RAS pathway mutations are observed primarily in tAML, it is plausible that the mutational variability between tMDS and tAML might have partly contributed to the opposing tAML and tMDS risk trends. In myeloma cohort the risk trends differed, with higher tAML risks over time and lower tMDS risk. Treatment with IMID has shown impact on CH genes (TET2, ASXL1, SRSF2 and SF3B1) 27 , and potentiation of TP53 clone, leading to increased tAML risk 28 .
Risk of both tMDS and tAML after early stage NSCLC (localized and regional) showed a steep increase in the second period (2006)(2007)(2008)(2009)(2010). A rapid adaptation of adjuvant chemotherapy with platinum agents in the treatment of early stage lung cancers during the first period (2000-2005) may directly explain this trend [29][30][31][32] . In the most recent period (2011-2016), we noted a decline in tAML risk after early stage (regional but not localized) NSCLC. Targeted therapies (EGFR/ALK-TKIs) and ICIs have revolutionized treatment of advanced NSCLC and about 25% of NSCLC patients can harbor EGFR and ALK mutations 9 , but these were not approved for early stage NSCLC during periods of our study. Therefore, decrease in tAML risk after regional stage NSCLC in the most recent period is rather surprising. It is expected that a high proportion of these early stage NSCLCs (especially, regional) have distant recurrences over time 33 and go on to receive either TKIs and ICIs that increase tumor surveillance. Localized NSCLCs have lower likelihood of distant recurrence 33 when compared to regional NSCLCs and hence, are less likely to be recipients of newer therapies. This would explain lack of downtrend in tAML risk after localized stages (Fig. 1).
Studies conducted thus far have been limited by the smaller number of observed tMDS/AML cases after melanoma and RCC or due to lack of receipt of chemotherapy [5][6][7] . For RCC, chemotherapy use was common until the approval of VEGF TKIs in 2005. With decreasing chemotherapy use and introduction of targeted therapies and ICIs, a potential downtrend in tAML risk is evident in 2011-2016 after RCC and melanoma; however, an uptrend is emerging for tMDS risk. Unique disease biology of tAML; for instance, FLT3 [an important tAML 18 and pre-tAML 4 clone] is a known target for sorafenib and sunitinib 34 and used extensively for RCC treatment 11 .
Historically, MM patients had the highest risk for tAML 5 . With decreasing use of alkylators, this risk was declining by the end of last century; however, more recently, the risk increased 5,35 , after the introduction of novel agents (e.g. IMiDs and proteasome inhibitors). Similar to previous SEER studies, we observed continued increase in tAML risk as stem cell transplantation and lenalidomide continue to remain a cornerstone of myeloma therapy 5,36 . A selective advantage of the RAS pathway gene mutations as well as TP53 clones 28,37,38 may have contributed to an increased risk for tAML after MM. It was interesting to note declining tMDS risk after MM. Recent epidemiological studies have noted that overall MDS incidence has been declining since 2011 39 , which may in part explain the trend noted. Alternatively, this observation may align with report on impact of IMiDs on CH 27 . Both these findings were consistent with our multivariate Poisson regression modeling which reaffirmed that risk for tAML after MM in patients initially treated with chemotherapy was significantly elevated (sTable 5); whereas, risk for tMDS did not differ by initial chemotherapy treatment.
We report new emerging trends of myeloid neoplasms since the advent of targeted and immunotherapies. In addition to above, it is key to consider possibility of surveillance bias in survivors of first primary cancers due to increased surveillance and monitoring of peripheral blood count. It is debatable if this would lead to distinctions observed in the type of secondary myeloid cancers. tMDS and tAML are often classified together 16 ; in our report we highlight their unique population trends. Limitations include the lack of data on specific chemotherapy, immunotherapy or targeted drugs and treatment patterns of recurrent/progressive disease. It is important to acknowledge that not all newly-diagnosed AML/MDS patients with prior solid malignancy including those www.nature.com/scientificreports/ with chemo-and radiotherapy exposure will have therapy-"related" disease. There likely are other factors at play such as germline predisposition, smoking etc., which are unaccounted in this study. Additionally, to make the time periods comparable, we truncated follow-up in the first two time periods to 5 years. The most recent time period had shorter follow-up, which could bias our results towards the null. However, if a bias existed, we would anticipate a similar trend for both tMDS and tAML.
In conclusion, we present unique and diverging trends of tMDS and tAML after a first primary cancer diagnosis in the era of modern cancer therapeutics. The underlying reasons are multifactorial and concordance of these trends with modern therapeutics is intriguing. With an increased understanding of the characteristics of CH and its evolution to tMDS/AML when exposed to external stressors, our future research focus remains on delineating the interaction of cancer therapies and pre-leukemic clones.

Data availability
All data are publicly available using SEER*Stat. Presented in part in the poster discussion session at the 2020 ASCO Virtual Scientific Program.