A simulation-based bias analysis to assess the impact of unmeasured confounding when designing non-randomized database studies.

BACKGROUND: Unmeasured confounding is often raised as a source of potential bias during the design of non-randomized studies but quantifying such concerns is challenging. METHODS: We developed a simulation-based approach to assess the potential impact of unmeasured confounding during the study design stage. The approach involved generation of hypothetical individual-level cohorts using realistic parameters including a binary treatment (prevalence 25%), a time-to-event outcome (incidence 5%), 13 measured covariates, a binary unmeasured confounder (u1, 10%), and a binary measured 'proxy' variable (p1) correlated with u1. Strength of unmeasured confounding and correlations between u1 and p1 were varied in simulation scenarios. Treatment effects were estimated with, a) no adjustment, b) adjustment for measured confounders (Level 1), c) adjustment for measured confounders and their proxy (Level 2). We computed absolute standardized mean differences in u1 and p1 and relative bias with each level of adjustment. RESULTS: Across all scenarios, Level 2 adjustment led to improvement in balance of u1, but this improvement was highly dependent on the correlation between u1 and p1. Level 2 adjustments also had lower relative bias than Level 1 adjustments (in strong u1 scenarios: relative bias of 9.2%, 12.2%, 13.5% at correlations 0.7, 0.5, and 0.3, respectively versus 16.4%, 15.8%, 15.0% for Level 1, respectively). CONCLUSION: An approach using simulated individual-level data was useful to explicitly convey the potential for bias due to unmeasured confounding while designing non-randomized studies and can be helpful in informing design choices.

The FAIRification of research in real-world evidence: A practical introduction to reproducible analytic workflows using Git and R.

Transparency and reproducibility are major prerequisites for conducting meaningful real-world evidence (RWE) studies that are fit for decision-making. Many advances have been made in the documentation and reporting of study protocols and results, but the principles for version control and sharing of analytic code in RWE are not yet as established as in other quantitative disciplines like computational biology and health informatics. In this practical tutorial, we aim to give an introduction to distributed version control systems (VCS) tailored toward the FAIR (Findable, Accessible, Interoperable, and Reproducible) implementation of RWE studies. To ease adoption, we provide detailed step-by-step instructions with practical examples on how the Git VCS and R programming language can be implemented into RWE study workflows to facilitate reproducible analyzes. We further discuss and showcase how these tools can be used to track changes, collaborate, disseminate, and archive RWE studies through dedicated project repositories that maintain a complete audit trail of all relevant study documents.

Deep Learning-based Propensity Scores for Confounding Control in Comparative Effectiveness Research: A Large-scale, Real-world Data Study.

BACKGROUND: Due to the non-randomized nature of real-world data, prognostic factors need to be balanced, which is often done by propensity scores (PSs). This study aimed to investigate whether autoencoders, which are unsupervised deep learning architectures, might be leveraged to compute PS. METHODS: We selected patient-level data of 128,368 first-line treated cancer patients from the Flatiron Health EHR-derived de-identified database. We trained an autoencoder architecture to learn a lower-dimensional patient representation, which we used to compute PS. To compare the performance of an autoencoder-based PS with established methods, we performed a simulation study. We assessed the balancing and adjustment performance using standardized mean differences, root mean square errors (RMSE), percent bias, and confidence interval coverage. To illustrate the application of the autoencoder-based PS, we emulated the PRONOUNCE trial by applying the trial's protocol elements within an observational database setting, comparing two chemotherapy regimens. RESULTS: All methods but the manual variable selection approach led to well-balanced cohorts with average standardized mean differences <0.1. LASSO yielded on average the lowest deviation of resulting estimates (RMSE 0.0205) followed by the autoencoder approach (RMSE 0.0248). Altering the hyperparameter setup in sensitivity analysis, the autoencoder approach led to similar results as LASSO (RMSE 0.0203 and 0.0205, respectively). In the case study, all methods provided a similar conclusion with point estimates clustered around the null (e.g., HRautoencoder 1.01 [95% confidence interval = 0.80, 1.27] vs. HRPRONOUNCE 1.07 [0.83, 1.36]). CONCLUSIONS: Autoencoder-based PS computation was a feasible approach to control for confounding but did not perform better than some established approaches like LASSO.

Changes in population-level survival for advanced solid malignancies with new treatment options in the second decade of the 21st century.

BACKGROUND: Several new treatments that improve survival in clinical trials have been developed for various solid malignancies in advanced stages. The effect of these options on survival in the general population is currently unknown. METHODS: Cancers for which 2 or more new treatment options have been approved by the US Food and Drug Administration during the years 2009 through 2011 for the treatment of advanced disease were identified, including adenocarcinoma of the lung, melanoma, breast cancer, prostate cancer, and renal cell carcinoma. Kaplan-Meier analysis was used to compare overall survival for these conditions in the Surveillance, Epidemiology, and End Results database for the periods 2007 to 2008, 2009 to 2010, and 2011 to 2012. Hazard ratios derived from adjusted, shared frailty models for cancer-specific survival were calculated as well for the years of diagnosis (2007-2008, 2009-2010, and 2011-2012). RESULTS: Two-year survival increased for patients with advanced-stage lung adenocarcinoma (+3.0 percentage points), melanoma (+3.4 percentage points), and breast cancer (+2.7 percentage points). When only patients aged 15 to 64 years were included, 2-year survival for those with melanoma increased by +6.7 percentage points. No change in survival was observed for renal cell carcinoma. Decreases in the hazard ratio for cancer-specific mortality were observed during the period 2011 to 2012 compared with 2007 to 2008 for lung adenocarcinoma, melanoma, and breast cancer. CONCLUSIONS: Small increases in 2-year survival were observed between the periods 2007 to 2008 and 2011 to 2012 for lung adenocarcinoma, melanoma, and prostate cancer. Cancer-specific mortality decreased for each of these cancers among patients who were diagnosed between the periods 2007 to 2008 and 2011 to 2013. These findings suggest that newer treatment options are beginning to increase survival for stage IV cancers at the population level.

Decreasing Use of Chemotherapy in Older Patients With Stage III Colon Cancer Irrespective of Comorbidities.

BACKGROUND: Chemotherapy underuse in elderly patients (aged ≥75 years) with colon cancer has been reported in previous studies. However, these studies were mostly registry-based and limited in their potential to consider underlying reasons of such undertreatment. This study aimed to evaluate patient and hospital determinants of chemotherapeutic treatment in patients with stage III colon cancer, with a particular focus on age and underlying reasons for nontreatment of elderly patients. METHODS: A total of 629 patients with stage III colon cancer who were diagnosed in 2003 through 2012 and recruited into a population-based study in the Rhine-Neckar region of Germany were included. Information on sociodemographic and lifestyle factors, comorbidities, and treatment was collected from patient interviews and physicians. Patient (with an emphasis on age) and hospital factors were evaluated for their associations with administration of adjuvant chemotherapy overall and of oxaliplatin specifically using multivariable logistic regression. RESULTS: Administration of chemotherapy decreased from 94% in patients aged 30 to 64 years to 51% in those aged ≥75 years. A very strong decline in chemotherapy use with age persisted even after comprehensive adjustment for multiple patient factors-including comorbidities-and hospital factors and was also seen among patients without any major comorbidities. Between 2005 and 2008, and 2009 and 2012, chemotherapy administration in patients aged ≥75 years decreased from 60% to 41%. Among chemotherapy recipients, old age was also strongly associated with higher odds of nonadministration of oxaliplatin. The 2 most commonly reported reasons for chemotherapy nonreceipt among the study population were patient refusal (30%) and old age (24%). CONCLUSIONS: Age was the strongest predictor of chemotherapy underuse, irrespective of comorbidities and even in patients without comorbidities. Such underuse due just to older age in otherwise healthy patients deserves increased attention in clinical practice to ensure that elderly patients also get the best possible care. Patients' refusal as the most frequent reason for chemotherapy nonreceipt also warrants further investigation to exclude misinformation as underlying cause.

A population-based registry study on relative survival from melanoma in Germany stratified by tumor thickness for each histologic subtype.

BACKGROUND: Differences in relative survival (RS) of melanoma between histologic subtypes were discussed to be mainly caused by tumor thickness. OBJECTIVE: To investigate RS of melanoma, stratified by tumor thickness for each histologic subtype, and identify survival trends. METHODS: With use of cancer registry data on melanoma cases (International Classification of Diseases, 10th Revision, codes C43.0-C43.9) diagnosed in Germany in 1997-2013, 5- and 10-year age-standardized RS stratified by histologic subtype and stratified or standardized by T stage was estimated by standard and modeled period analyses. We restricted 10-year RS analyses to patients younger than 75 years. RESULTS: We analyzed 82,901 cases. Overall, the 5- and 10-year RS rates were 91.7% and 90.8%, respectively. Prognosis worsened with increasing T stage for all histologic subtypes, but T-stage distribution varied substantially. Survival differences by histologic subtype were strongly alleviated after adjustment for T stage but remained significant. Overall, 5-year RS increased significantly (by 3.8 percentage points) between the periods 2002-2005 and 2010-2013. This increase was no longer seen after adjustment for T stage. LIMITATIONS: Exclusion of cases on account of missing information on T stages, changes in the definition of T stages, and lack of information on screening and treatment limit our analyses. CONCLUSION: Differences in RS between histologic subtypes were strongly mediated by tumor thickness. Over time, RS of melanoma increased as a result of changes in T-stage distribution.

Long-term heart-specific mortality among 347 476 breast cancer patients treated with radiotherapy or chemotherapy: a registry-based cohort study.

Aims: Breast cancer survival has improved throughout the last decades, but treatment-induced cardiotoxicity remains a major concern. This study aimed to investigate competing causes of death and prognostic factors within a large cohort of breast cancer patients and to describe the heart-specific mortality in relation to the general population. Methods and results: In this registry-based cohort study, women diagnosed with breast cancer between 2000 and 2011, who were treated with radiotherapy or chemotherapy and followed until 2014, were identified from the Surveillance, Epidemiology, and End Results-18 (SEER-18) database. Cumulative mortality functions were computed. To investigate heart-specific mortality relative to the general population, long-term (≥10 years) standardized mortality ratios (SMRs) were calculated. Prognostic factors for heart-specific mortality were assessed by calculating cause-specific hazard ratios (HRcs) with corresponding 95% confidence intervals using the Cox proportional hazards regression. Subgroup analysis on intermediate-term mortality according to molecular subtypes, for which information was available since 2010, was performed. In total, 347 476 breast cancer patients were eligible to be included in the study. Among all possible competing causes of death, breast cancer accounted for the highest cumulative mortality. Compared with the general population, heart-specific mortality of breast cancer patients treated with radiotherapy or chemotherapy was lower [SMRoverall 0.84 (0.79-0.90)]. In subgroup analysis, human epidermal growth factor receptor 2 (HER2)-positive subtype was not associated with increased heart-specific mortality relative to HER2-negative patients [HRcs 0.96 (0.70-1.32)]. Conclusion: Heart-specific mortality among breast cancer survivors is not increased compared with the general population. Human epidermal growth factor receptor 2-positive patients do not have increased heart-specific mortality compared to HER2-negative patients.

Survival of patients with hepatobiliary tract and duodenal cancer sites in Germany and the United States in the early 21st century.

Hepatobiliary tract cancers (HBTCs) are a heterogeneous group of cancers with high mortality. Because most of these cancers, with the exception of hepatocellular carcinoma (HCC), are rare, few data are available concerning the population level survival expectations of patients with HBTC. Here, we describe survival of patients with HBTC in Germany with comparison to survival in the US. Therefore, data were extracted from 12 databases in Germany and the Surveillance, Epidemiology and End Results (SEER13) database in the US. Period analysis and modeled period analysis were used to calculate 5-year relative survival estimates for patients with HBTC diagnosed from 1997 to 2013. HCC was the most common HBTC in each database, accounting for over 1/3 of HBTC in Germany and about half of cases in the US. Overall age adjusted 5-year relative survival for HBTC in 2006-2013 was 19.1% in Germany and 20.6% in the US. Five-year relative survival increased by 3.8% units in Germany and 4.5% units in the US between 2002-2005 and 2010-2013. Five-year relative survival for individual types of HBTC ranged from 9.8% in Germany and 2.9% in the US for not otherwise specified biliary tract cancers to 44.4% and 50.1%, respectively, in Germany and the US for duodenal cancers. In conclusion, survival for HBTC remains poor in both Germany and the US, although a small increase in survival in the past decade was observed. Further work to find better treatment options for HBTC is needed to improve survival.

Pre- and post-diagnostic beta-blocker use and prognosis after colorectal cancer: Results from a population-based study.

Recent experimental and epidemiological studies have suggested that beta-blocker use might be associated with better cancer prognosis, but results were inconclusive and only few studies have investigated the association specifically for colorectal cancer (CRC) patients. We investigated this hypothesis using a linked dataset of the Eindhoven area of the Netherlands Cancer Registry and the PHARMO record linkage, including patients diagnosed with CRC between 1998 and 2011. Eligible patients were matched on propensity scores to control for potential confounders such as socio-demographic factors, comorbidity, cancer treatment and use of other medications. Controls were subsequently restricted to active comparators. The association between pre-diagnostic and time-dependent post-diagnostic beta-blocker use and overall survival was estimated using Cox proportional hazard regression models. Subgroup analyses by cancer site and stage and by beta-blocker type were conducted. Of 8,100 CRC patients with a median follow-up of 6.6 years, 1,813 (22%) used beta-blockers prior to diagnosis. In multivariate analysis, we observed no significant association in overall mortality for pre-diagnostic [hazard ratio 1.07, 95% confidence interval (0.96-1.19)] and post-diagnostic [1.10 (0.98-1.23)] beta-blocker use, respectively. Analyses by beta-blocker type, by cancer site, cancer stage and by cumulative dose showed no significant survival improvements for beta-blocker users. However, there was a significant association between cumulative duration of use of 1-12 months and increased overall mortality [1.20 (1.03-1.39)]. Thus, our results do not support the hypothesis of a beneficial effect of pre- or post-diagnostic beta-blocker intake on CRC prognosis, neither for specific patient subgroups nor for specific types of beta-blockers.

Immortal time bias in pharmacoepidemiological studies on cancer patient survival: empirical illustration for beta-blocker use in four cancers with different prognosis.

Immortal time bias (ITB) is still seen frequently in medical literature. However, not much is known about this bias in the field of cancer (pharmaco-)epidemiology. In context of a hypothetical beneficial beta-blocker use among cancer patients, we aimed to demonstrate the magnitude of ITB among 9876 prostate, colorectal, lung and pancreatic cancer patients diagnosed between 1998 and 2011, which were selected from a database linkage of the Netherlands Cancer Registry and the PHARMO Database Network. Hazard ratios (HR) and 95% confidence intervals from three ITB scenarios, defining exposure at a defined point after diagnosis (model 1), at any point after diagnosis (model 2) and as multiple exposures after diagnosis (model 3), were calculated to investigate the association between beta-blockers and cancer prognosis using Cox proportional hazards regression. Results were compared to unbiased estimates derived from the Mantel-Byar model. Ignoring ITB led to substantial smaller HRs for beta-blocker use proposing a significant protective association in all cancer types [e.g. HR 0.18 (0.07-0.43) for pancreatic cancer in model 1], whereas estimates derived from the Mantel-Byar model were mainly suggesting no association [e.g. HR 1.10 (0.84-1.44)]. The magnitude of bias was consistently larger among cancer types with worse prognosis [overall median HR differences between all scenarios in model 1 and Mantel-Byar model of 0.56 (prostate), 0.72 (colorectal), 0.77 (lung) and 0.85 (pancreas)]. In conclusion, ITB led to spurious beneficial associations of beta-blocker use among cancer patients. The magnitude of ITB depends on the duration of excluded immortal time and the prognosis of each cancer.

smdi: an R package to perform structural missing data investigations on partially observed confounders in real-world evidence studies.

Objectives: Partially observed confounder data pose a major challenge in statistical analyses aimed to inform causal inference using electronic health records (EHRs). While analytic approaches such as imputation are available, assumptions on underlying missingness patterns and mechanisms must be verified. We aimed to develop a toolkit to streamline missing data diagnostics to guide choice of analytic approaches based on meeting necessary assumptions. Materials and methods: We developed the smdi (structural missing data investigations) R package based on results of a previous simulation study which considered structural assumptions of common missing data mechanisms in EHR. Results: smdi enables users to run principled missing data investigations on partially observed confounders and implement functions to visualize, describe, and infer potential missingness patterns and mechanisms based on observed data. Conclusions: The smdi R package is freely available on CRAN and can provide valuable insights into underlying missingness patterns and mechanisms and thereby help improve the robustness of real-world evidence studies.

A Principled Approach to Characterize and Analyze Partially Observed Confounder Data from Electronic Health Records.

Objective: Partially observed confounder data pose challenges to the statistical analysis of electronic health records (EHR) and systematic assessments of potentially underlying missingness mechanisms are lacking. We aimed to provide a principled approach to empirically characterize missing data processes and investigate performance of analytic methods. Methods: Three empirical sub-cohorts of diabetic SGLT2 or DPP4-inhibitor initiators with complete information on HbA1c, BMI and smoking as confounders of interest (COI) formed the basis of data simulation under a plasmode framework. A true null treatment effect, including the COI in the outcome generation model, and four missingness mechanisms for the COI were simulated: completely at random (MCAR), at random (MAR), and two not at random (MNAR) mechanisms, where missingness was dependent on an unmeasured confounder and on the value of the COI itself. We evaluated the ability of three groups of diagnostics to differentiate between mechanisms: 1)-differences in characteristics between patients with or without the observed COI (using averaged standardized mean differences [ASMD]), 2)-predictive ability of the missingness indicator based on observed covariates, and 3)-association of the missingness indicator with the outcome. We then compared analytic methods including "complete case", inverse probability weighting, single and multiple imputation in their ability to recover true treatment effects. Results: The diagnostics successfully identified characteristic patterns of simulated missingness mechanisms. For MAR, but not MCAR, the patient characteristics showed substantial differences (median ASMD 0.20 vs 0.05) and consequently, discrimination of the prediction models for missingness was also higher (0.59 vs 0.50). For MNAR, but not MAR or MCAR, missingness was significantly associated with the outcome even in models adjusting for other observed covariates. Comparing analytic methods, multiple imputation using a random forest algorithm resulted in the lowest root-mean-squared-error. Conclusion: Principled diagnostics provided reliable insights into missingness mechanisms. When assumptions allow, multiple imputation with nonparametric models could help reduce bias.

A systematic approach towards missing lab data in electronic health records: A case study in non-small cell lung cancer and multiple myeloma.

Real-world data derived from electronic health records often exhibit high levels of missingness in variables, such as laboratory results, presenting a challenge for statistical analyses. We developed a systematic workflow for gathering evidence of different missingness mechanisms and performing subsequent statistical analyses. We quantify evidence for missing completely at random (MCAR) or missing at random (MAR), mechanisms using Hotelling's multivariate t-test, and random forest classifiers, respectively. We further illustrate how to apply sensitivity analyses using the not at random fully conditional specification procedure to examine changes in parameter estimates under missing not at random (MNAR) mechanisms. In simulation studies, we validated these diagnostics and compared analytic bias under different mechanisms. To demonstrate the application of this workflow, we applied it to two exemplary case studies with an advanced non-small cell lung cancer and a multiple myeloma cohort derived from a real-world oncology database. Here, we found strong evidence against MCAR, and some evidence of MAR, implying that imputation approaches that attempt to predict missing values by fitting a model to observed data may be suitable for use. Sensitivity analyses did not suggest meaningful departures of our analytic results under potential MNAR mechanisms; these results were also in line with results reported in clinical trials.

The Impact of Longitudinal Data-Completeness of Electronic Health Record Data on the Prediction Performance of Clinical Risk Scores.

The impact of electronic health record (EHR) discontinuity (i.e., receiving care outside of a given EHR system) on EHR-based risk prediction is unknown. We aimed to assess the impact of EHR-continuity on the performance of clinical risk scores. The study cohort consisted of patients aged ≥ 65 years with ≥ 1 EHR encounter in the 2 networks in Massachusetts (MA; 2007/1/1-2017/12/31, internal training and validation dataset), and one network in North Carolina (NC; 2007/1/1-2016/12/31, external validation dataset) that were linked with Medicare claims data. Risk scores were calculated using EHR data alone vs. linked EHR-claims data (not subject to misclassification due to EHR-discontinuity): (i) combined comorbidity score (CCS), (ii) claim-based frailty score (CFI), (iii) CHAD2 DS2 -VASc, and (iv) Hypertension, Abnormal renal/liver function, Stroke, Bleeding, Labile, Elderly, and Drugs (HAS-BLED). We assessed the performance of CCS and CFI predicting death, CHAD2 DS2 -VASc predicting ischemic stroke, and HAS-BLED predicting bleeding by area under receiver operating characteristic curve (AUROC), stratified by quartiles of predicted EHR-continuity (Q1-4). There were 319,740 patients in the MA systems and 125,380 in the NC system. In the external validation dataset, AUROC for EHR-based CCS predicting 1-year risk of death was 0.583 in Q1 (lowest) EHR-continuity group, which increased to 0.739 in Q4 (highest) EHR-continuity group. The corresponding improvement in AUROC was 0.539 to 0.647 for CFI, 0.556 to 0.637 for CHAD2 DS2 -VASc, and 0.517 to 0.556 for HAS-BLED. The AUROC in Q4 EHR-continuity group based on EHR alone approximates that based on EHR-claims data. The prediction performance of four clinical risk scores was substantially worse in patients with lower vs. high EHR-continuity.

Clinical outcomes of COVID-19 in patients with multiple sclerosis treated with ocrelizumab in the pre- and post-SARS-CoV-2 vaccination periods: Insights from Israel.

The coronavirus disease 2019 (COVID-19) pandemic caused challenges in the management of patients living with multiple sclerosis (PLwMS). We investigated the occurrence and severity of COVID-19 infection post-vaccination among PLwMS treated with ocrelizumab and enrolled in the Maccabi Health Services (MHS) (n = 289) or followed at the Hadassah Medical Center (HMC) (n = 80) in Israel. Most patients were fully vaccinated (MHS n = 218; HMC n = 76) and confirmed infection post-vaccination was low (3.7% and 2.6%, respectively). MHS: infection was more severe (hospitalization/intensive care unit/death) in non-vaccinated (33.3%) vs vaccinated patients (25%). HMC: one vaccinated patient required hospitalization with COVID-19 vs two unvaccinated patients. These data from two Israel cohorts suggest that occurrence of COVID-19 after mRNA vaccination is low and limited in severity.

Comparative Effectiveness of Immune Checkpoint Inhibitors vs Chemotherapy by Tumor Mutational Burden in Metastatic Castration-Resistant Prostate Cancer.

Importance: The most useful biomarkers for clinical decision-making identify patients likely to have improved outcomes with one treatment vs another. Objective: To evaluate treatment class-specific outcomes of patients receiving immune checkpoint inhibitor (ICI) vs taxane chemotherapy by tumor mutational burden (TMB). Design, Setting, and Participants: This comparative effectiveness analysis of clinical variables and outcomes used prospectively defined biomarker-stratified genomic data from a deidentified clinicogenomic database. Data included men with previously treated metastatic castration-resistant prostate cancer (mCRPC) receiving ICI or single-agent taxane chemotherapy from January 2011 to April 2021 at approximately 280 US academic or community-based cancer clinics (approximately 800 sites of care). Data were analyzed from July to August 2021. Exposures: Single-agent ICI or single-agent taxanes. Treatments were assigned at discretion of physician and patient without randomization. Imbalances of known factors between treatment groups were adjusted with propensity weighting. Main Outcomes and Measures: Prostate-specific antigen (PSA) response, time to next therapy (TTNT), and overall survival (OS). Results: A total of 741 men (median [IQR], 70 [64-76] years) with mCRPC received comprehensive genomic profiling and were treated with ICI or single-agent taxane therapy. At baseline, the median (IQR) PSA level was 79.4 (19.0-254) ng/mL, 108 men (18.8%) had Eastern Cooperative Oncology Group Performance Status scores of 2 or greater, and 644 men (86.9%) had received prior systemic treatments for mCRPC. A total of 45 patients (6.1%) received ICI therapy and 696 patients (93.9%) received taxane therapy. Among patients with TMB of fewer than 10 mutations per megabase (mt/Mb) receiving ICI, compared with those receiving taxanes, had worse TTNT (median [IQR], 2.4 [1.1-3.2] months vs 4.1 [2.2-6.3] months; hazard ratio [HR], 2.65; 95% CI, 1.78-3.95; P < .001). In contrast, for patients with TMB of 10 mt/Mb or greater, use of ICIs, compared with use taxanes, was associated with more favorable TTNT (median [IQR], 8.0 [3.4 to unknown] months vs 2.4 [2.4-7.3] months; HR, 0.37, 95% CI, 0.15-0.87; P = .02) and OS (median 19.9 [8.06 to unknown] months vs 4.2 [2.69 - 6.12] months; HR, 0.23; 95% CI, 0.10-0.57; P = .001). Among all 741 patients, 44 (5.9%) had TMB of 10 mt/Mb or greater, 22 (3.0%) had high microsatellite instability, and 20 (2.7%) had both. Treatment interactions with TMB of 10 mt/Mb or greater (TTNT: HR, 0.10; 95% CI, 0.32-0.31; P < .001; OS: HR, 0.25; 95% CI, 0.076-0.81; P = .02) were stronger than high microsatellite instability alone (TTNT: HR, 0.12; 95% CI, 0.03-0.51; P = .004; OS: HR, 0.38; 95% CI, 0.13-1.12; P = .08). Conclusions and Relevance: In this comparative effectiveness study, ICIs were more effective than taxanes in patients with mCRPC when TMB was 10 mt/Mb or greater but not when TMB was fewer than 10 mt/Mb. The results add validity to the existing TMB cutoff of 10 mt/Mb for ICI use in later lines of therapy, and suggest that ICIs may be a viable alternative to taxane chemotherapy for patients with mCRPC with high TMB.

Artificial Intelligence for Prognostic Scores in Oncology: a Benchmarking Study.

Introduction: Prognostic scores are important tools in oncology to facilitate clinical decision-making based on patient characteristics. To date, classic survival analysis using Cox proportional hazards regression has been employed in the development of these prognostic scores. With the advance of analytical models, this study aimed to determine if more complex machine-learning algorithms could outperform classical survival analysis methods. Methods: In this benchmarking study, two datasets were used to develop and compare different prognostic models for overall survival in pan-cancer populations: a nationwide EHR-derived de-identified database for training and in-sample testing and the OAK (phase III clinical trial) dataset for out-of-sample testing. A real-world database comprised 136K first-line treated cancer patients across multiple cancer types and was split into a 90% training and 10% testing dataset, respectively. The OAK dataset comprised 1,187 patients diagnosed with non-small cell lung cancer. To assess the effect of the covariate number on prognostic performance, we formed three feature sets with 27, 44 and 88 covariates. In terms of methods, we benchmarked ROPRO, a prognostic score based on the Cox model, against eight complex machine-learning models: regularized Cox, Random Survival Forests (RSF), Gradient Boosting (GB), DeepSurv (DS), Autoencoder (AE) and Super Learner (SL). The C-index was used as the performance metric to compare different models. Results: For in-sample testing on the real-world database the resulting C-index [95% CI] values for RSF 0.720 [0.716, 0.725], GB 0.722 [0.718, 0.727], DS 0.721 [0.717, 0.726] and lastly, SL 0.723 [0.718, 0.728] showed significantly better performance as compared to ROPRO 0.701 [0.696, 0.706]. Similar results were derived across all feature sets. However, for the out-of-sample validation on OAK, the stronger performance of the more complex models was not apparent anymore. Consistently, the increase in the number of prognostic covariates did not lead to an increase in model performance. Discussion: The stronger performance of the more complex models did not generalize when applied to an out-of-sample dataset. We hypothesize that future research may benefit by adding multimodal data to exploit advantages of more complex models.

Divergent Patterns and Trends in Breast Cancer Incidence, Mortality and Survival Among Older Women in Germany and the United States.

BACKGROUND: Breast cancer treatment has changed tremendously over the last decades. In addition, the use of mammography screening for early detection has increased strongly. To evaluate the impact of these developments, long-term trends in incidence, mortality, stage distribution and survival were investigated for Germany and the United States (US). METHODS: Using population-based cancer registry data, long-term incidence and mortality trends (1975-2015), shifts in stage distributions (1998-2015), and trends in five-year relative survival (1979-2015) were estimated. Additionally, trends in five-year relative survival after standardization for stage were explored (2004-2015). RESULTS: Age-standardized breast cancer incidence rates were much higher in the US than in Germany in all periods, whereas age-standardized mortality began to lower in the US from the 1990s on. The largest and increasing differences were observed for patients aged 70+ years with a 19% lower incidence but 45% higher mortality in Germany in 2015. For this age group, large differences in stage distributions were observed, with 29% (Germany) compared to 15% (US) stage III and IV patients. Age-standardized five-year relative survival increased strongly between 1979-1983 and 2013-2015 in Germany (+17% units) and the US (+19% units) but was 9% units lower in German patients aged 70+ years in 2013-2015. This difference was entirely explained by differences in stage distributions. CONCLUSIONS: Overall, our results are in line with a later uptake and less extensive utilization of mammography screening in Germany. Further studies and efforts are highly needed to further explore and overcome the increased breast cancer mortality among elderly women in Germany.

Long-term relative survival from melanoma in Germany 1997-2013.

Up-to-date melanoma relative survival (RS) estimates and trend analysis facilitate close monitoring of melanoma patients' prognosis. This study aimed to provide recent 5-year and 10-year RS from melanoma, stratified by prognostic factors, and identify latest survival trends. Data from 12 German cancer registries were analysed. We included patients with primary cutaneous malignant melanoma (ICD-10: C43.X) diagnosed in 1997-2013 who were at least 15 years old. Five-year and 10-year RS were estimated by period analysis. For 10-year RS analyses, we excluded patients who were 75 years of age or older. Analyses were stratified by sex, age, histology, tumour stage, and body site. We included 82 901 patients, of whom 51% were women. The median age at diagnosis was 62 years. Five-year and 10-year RS in 2007-2013 were 92.4 and 90.8%, respectively. RS was higher in women. The prognosis worsened with older age and higher stage. In superficial spreading melanoma and lentigo maligna melanoma, RS was high; it was lower in nodular, acral lentiginous and 'other' melanoma. RS was the highest for melanoma on the arms; RS for melanoma on unknown or overlapping sites of the skin was the lowest. Five-year and 10-year RS increased significantly from 2005-2007 and 2008-2010 to 2011-2013, by 3.5 and 3.3 percentage points, respectively. For melanoma of 'other' histology, 5-year and 10-year RS increased significantly. Ten-year RS also increased significantly in men with superficial spreading melanoma and T4 melanoma, and in women with T3 melanoma. Melanoma RS improved, especially in certain subgroups. The reasons for improvements need to be investigated further.