Abstract
Background
Limited data compare first-line carboplatin-based chemotherapy and immune checkpoint blockade in cisplatin-ineligible metastatic urothelial carcinoma (mUC) patients. The primary evidence guiding treatment decisions was a recent Food and Drug Administration/European Medicines Agency safety alert based on emerging data from two ongoing phase III trials, reporting shorter survival in programmed death-ligand 1 (PD-L1) negative patients receiving immunotherapy. Final results from these trials are unknown.
Objective
To compare survival in cisplatin-ineligible mUC patients receiving first-line immunotherapy versus carboplatin-based chemotherapy.
Design, Setting, and Participants
We conducted a retrospective cohort study of 2,017 mUC patients receiving first-line carboplatin-based chemotherapy (n=1,530) or immunotherapy (n=487) from January 1, 2011 to May 18, 2018 using the Flatiron Health electronic health record-derived database.
Outcome Measurements and Statistical Analysis
The primary outcome was overall survival (OS), comparing 12- and 36-month OS, and hazard ratios before and after 12 months. Propensity score-based inverse probability of treatment weighting (IPTW) was used to address confounding in Kaplan-Meier and Cox regression model estimates of comparative effectiveness.
Results and Limitations
IPTW-adjusted OS rates in the immunotherapy group were lower at 12 months (39.6% [95% CI 34.0-45.3%] versus 46.1% [95% CI 43.4-48.8%]) but higher at 36 months (28.3% [95% CI 21.8-34.7%] versus 13.3% [95% CI 11.1-15.5%]) relative to the chemotherapy group. Immunotherapy treatment demonstrated inferior OS during the first 12 months relative to carboplatin-based chemotherapy (IPTW-adjusted HR 1.37, 95% CI 1.15-1.62), but superior OS beyond 12 months (IPTW-adjusted HR 0.50, 95% CI 0.30-0.85). Limitations include retrospective design and potential unmeasured confounding.
Conclusions
In the setting of mUC, clinicians and patients should carefully consider how to balance the short-term benefit of chemotherapy against the long-term benefit of immunotherapy.
Patient Summary
To determine the optimal first-line therapy for metastatic bladder cancer patients who are unfit for cisplatin, we compared carboplatin-based chemotherapy versus immunotherapy using real-world data. Survival in the first year of treatment was lower with immunotherapy relative to chemotherapy, but for patients surviving beyond the first year, immunotherapy was superior.
Keywords: EMA, FDA, immune checkpoint blockade, metastatic urothelial cancer, real-world data
Introduction
Metastatic urothelial carcinoma primarily affects older individuals. As a result, age-related comorbidity precludes over 50% of patients from receiving standard cisplatin-based chemotherapy – the only first-line treatment shown to improve survival.1 For cisplatin-ineligible patients, outcomes are poor and there is no universally accepted treatment standard. Historically, carboplatin-based regimens have been used in this setting.2,3,4
Recently, two immune checkpoint inhibitors, pembrolizumab and atezolizumab, received accelerated approval for front-line use in cisplatin-ineligible patients, providing an alternative to carboplatin-based chemotherapy. However, no available data directly comparing these first-line treatment strategies. Expedited approval of immunotherapy was based on surrogate endpoints (e.g., response rates) from two phase II single-arm trials KEYNOTE-052 and IMVigor-210.5,6,6 Response rates from these uncontrolled immunotherapy trials (~24-29%) were lower than those seen in trials of carboplatin-based chemotherapy (~40-45%).1,3,4 Without comparative data using patient-centered endpoints (e.g., survival), important effectiveness information may be missed, preventing informed decision-making. Further complicating treatment selection, recently, the Food and Drug Administration (FDA) and European Medicines Agency (EMA) issued a safety alert reporting decreased survival in programmed death-ligand 1 (PD-L1) negative mUC patients treated with immunotherapy relative to platinum-based chemotherapy.7,8 As a result, immunotherapy use was restricted to cisplatin-ineligible mUC patients who are PD-L1 positive or who are ineligible for any platinum-containing chemotherapy.9 Since the EMA and FDA reports were based on early review of two ongoing phase III trials of platinum-eligible patients, KEYNOTE-361 and IMVigor-130, the full results are unknown and applicability to routine clinical practice is uncertain.
While randomized clinical trial data represent the gold standard for therapeutic approvals, there is growing interest by patients, physicians, and regulators in leveraging real-world evidence to better inform practice, as emphasized in the 21st Century Cures Act.10 In this retrospective cohort study, we compared the effectiveness of immunotherapy versus carboplatin-based chemotherapy as first-line therapy for cisplatin-ineligible mUC patients in routine clinical practice.
Material and Methods
Reporting follows recommendations from the International Society for Pharmacoeconomics and Outcomes Research (ISPOR) and the International Society for Pharmacoepidemiology (ISPE) Special Task Force on Real World Evidence in Health Care Decision Making.11 The study protocol was approved by the University of Pennsylvania institutional review-board with waiver of informed consent.
Data Source
Data were obtained from the Flatiron Health electronic health record (EHR)-derived database, a geographically diverse United States database comprised of patient-level structured and unstructured data, curated via technology-enabled abstraction. All data from unstructured EHR-derived digital documents were manually reviewed by centrally-managed and trained medical record abstractors using explicit abstraction protocols for each data element.12 Quality control during the abstraction process consists of duplicate chart abstraction, logic checks, and formal adjudication based on the complexity of select variables, as has been described in previous analyses.13-15 The database includes de-identified data from over 280 academic and community oncology practices (~800 sites of care) representing more than 2.1 million United States cancer patients available for analysis. The data were cut by Flatiron Health on September 30, 2018 with recency from January 1, 2011 through August 31, 2018. The cohort is similar in age, race, and gender to the United States population with advanced urothelial carcinoma based on Surveillance, Epidemiology, and End Results (SEER) data from 2004-2013.16
Study Population
The study sample included patients diagnosed with stage IV urothelial carcinoma (bladder, renal pelvis, ureter, or urethra) and those diagnosed with early stage urothelial carcinoma who subsequently developed metastatic disease, and initiated first-line therapy (Supplemental Figure 1). Each patient had an ICD code for urothelial cancer, at least 2 documented clinical visits on or after January 1, 2011, pathology consistent with urothelial cell carcinoma, and confirmation of node-positive or metastatic disease between January 1, 2011 and May 18, 2018. The study excluded patients who did not receive systemic therapy for advanced bladder cancer, received first-line agents as part of a clinical trial, had greater than a 90-day gap between diagnosis and first structured EHR activity, or received first-line agents that were not listed in the National Comprehensive Cancer Network (NCCN) guidelines for systemic therapy of mUC. Patients were excluded if they initiated first-line treatment after May 18, 2018, the time of the FDA safety alert and subsequent label revision for mUC immunotherapy, to reduce confounding from choice of first-line therapy by PD-L1 status.
Exposure
Carboplatin-based chemotherapy was defined as a NCCN-guideline recommended carboplatin-containing doublet or other evidence-based carboplatin-containing regimen. Immunotherapy treatment was defined as single-agent nivolumab, pembrolizumab, atezolizumab, durvalumab, or avelumab.
Outcome Measures
The primary outcome was overall survival (OS), defined as the time from the start of the first-line treatment of interest to the date of death. Follow up was terminated at the earliest of death, data extraction date of August 31, 2018, or last activity in the EHR. The secondary outcome was second-line therapy-free survival defined as the time from the start of first-line therapy to the earliest of start date of second line therapy or death.
Mortality information in the Flatiron Health database is derived from structured and unstructured documents within the EHR, as well as the Social Security Death Index and a commercial death dataset that mines data from obituaries and funeral homes. When compared to the National Death Index, Flatiron Health’s mortality data showed high sensitivity (85-90%), specificity (97%), PPV (>96%), and exact date agreement (> 90%).17
Covariates
Eleven covariates thought likely to influence the choice of first-line therapy were measured in the 62 days prior to treatment start. These included patient factors such as age, sex, race/ethnicity, smoking status, body mass index (BMI), primary site of disease, Eastern Cooperative Oncology Group (ECOG) performance status (using the most proximate value to first treatment episode), comorbidity score (using ICD-9/10 diagnosis codes as outlined by Elixhauser et al.18), and use of opioid pain medication or corticosteroids as a surrogate of symptomatic or high-volume disease; as well as practice-factors including academic or community practice and immunotherapy prescribing rate by practice.
Statistical Analysis
Cohort characteristics in chemotherapy and immunotherapy groups were compared using standard descriptive statistics. Multiple imputation via chained equations was used to address missing data that were assumed to be missing at random.19 Ten imputed data sets were created, including all covariates listed above in the imputation model. Rubin’s rules were used to generate pooled effect estimates and variances across imputed data sets.20,21
To address systematic differences between chemotherapy and immunotherapy initiators (i.e., confounding by indication), we used inverse probability of treatment weighted (IPTW) analyses. IPTW, a form of propensity score analysis, uses weighting by the inverse of the propensity score to reduce imbalance in measured confounders between treatment groups.22 The propensity score model included all baseline characteristics listed above.
Propensity scores were estimated using Super Learner, an ensemble machine-learning algorithm.23 The algorithm combined weighted estimates across several parametric and nonparametric prediction modeling approaches based on the accuracy of the predictions from the models to create an overall propensity score estimate, which increased the robustness of the analysis. These estimated propensity scores were used to calculate each patient’s inverse probability of being treated with carboplatin-based chemotherapy or immunotherapy. Post-weighting balance in covariates between treatment groups was evaluated using the standardized differences approach. Imbalance was defined as a standardized difference greater than 10%. Overlap of propensity score distributions between treatment groups was assessed graphically using density plots (Supplemental Figure 2).
IPTW-adjusted Kaplan-Meier curves compared overall survival (median OS, 12-month OS, 36-month OS) between treatment groups. Multivariable Cox proportional hazards regression analysis estimated IPTW-adjusted hazard ratios (HRs) and 95% confidence intervals (CIs) for immunotherapy compared to carboplatin-based chemotherapy. The proportional hazards assumption was evaluated by testing the correlation of the scaled Schoenfeld residuals and time. After observing deviations from proportionality for the immunotherapy effect, we incorporated a time-varying coefficient for immunotherapy, allowing for a single change-point in the immunotherapy effect at 12 months. In a sensitivity analysis to assess the degree of unmeasured confounding, we calculated E-values based on point estimates before and after 12 months, to quantify the minimum strength of association between an unmeasured confounder and both the treatment and outcome needed to nullify the observed treatment-outcome association.24 In an exploratory analysis, we repeated the primary analysis, stratified by PD-L1 test status. PD-L1 was primarily assessed by immunohistochemistry using the Dako 22C3 or 28-8 assay (positive if combined positive score of ≥10%) or Ventana assay SP142 or SP263 (positive if tumor-infiltrating immune cells ≥5%). All statistical tests were two-sided, conducted at the 5% significance level, using R version 3.5.1.
Results
Unweighted and Weighted Baseline Characteristics
Of 2,017 patients, 487 patients received immunotherapy and 1,530 received carboplatin-based chemotherapy (Table 1). The median age was 78 years and the majority of patients were male (73%), white (74%), had a history of smoking (72%), and received treatment at a community practice (97%). Notably, PD-L1 was tested in only 7% of patients, consistent with the original label indication for immunotherapy which did not mandate PD-L1 testing for cisplatin-ineligible patients.
Table 1:
Baseline characteristics of patients (n=2,017) who received first-line immunotherapy or carboplatin-based chemotherapy
| Unweighted Population | Inverse Probability of Imputed Treatment Weighted Population |
|||||
|---|---|---|---|---|---|---|
| Immunotherapy n=487 |
Carboplatin- based chemotherapy n=1,530 |
SMD | Immunotherapy n=487 |
Carboplatin- based chemotherapy n=1,530 |
SMD | |
| Median age | 77 | 78 | 0.038 | 77 | 78 | 0.059 |
| Sex | ||||||
| Male | 360 (74%) | 1115 (73%) | 0.024 | 75% | 74% | 0.031 |
| Female | 127 (26%) | 415 (27%) | 25% | 26% | ||
| Race / Ethnicity | ||||||
| White | 358 (74%) | 1137 (74%) | 0.111 | 74% | 74% | 0.076 |
| Black | 18 (3.7%) | 75 (4.9%) | 4.5% | 4.9% | ||
| Other | 53 (11%) | 180 (12%) | 11% | 12% | ||
| Unknown | 58 (12%) | 138 (9.0%) | 11% | 9.2% | ||
| Year of diagnosis | ||||||
| 2011 | 0 (0%) | 109 (7%) | 2.081 | 0% | 6.7% | 1.923 |
| 2012 | 0 (0%) | 166 (11%) | 0% | 10% | ||
| 2013 | 0 (0%) | 219 (14%) | 0% | 14% | ||
| 2014 | 0 (0%) | 226 (15%) | 0% | 14% | ||
| 2015 | 8 (1.6%) | 277 (18%) | 2.4% | 18% | ||
| 2016 | 93 (19%) | 296 (19%) | 21% | 20% | ||
| 2017 | 265 (54%) | 184 (12%) | 55% | 13% | ||
| 2018 | 121 (25%) | 53 (3.5%) | 22% | 3.8% | ||
| Site of disease | ||||||
| Bladder | 367 (75%) | 1131 (74%) | 0.160 | 76% | 74% | 0.087 |
| Renal Pelvis | 63 (13%) | 256 (17%) | 13% | 16% | ||
| Ureter | 56 (12%) | 131 (8.6%) | 10% | 9.0% | ||
| Urethra | 1 (0.21%) | 12 (0.78%) | 0.47% | 0.71% | ||
| Smoking history | ||||||
| Current/former smoker | 362 (74%) | 1093 (71%) | 0.115 | 76% | 72% | 0.123 |
| Never smoker | 120 (25%) | 400 (26%) | 23% | 26% | ||
| Unknown | 5 (1.0%) | 37 (2.4%) | 0.91% | 2.2% | ||
| Mean BMI | 26 | 27 | 0.109 | 27 | 27 | 0.032 |
| ECOG performance status+ | ||||||
| 0 | 88 (26%) | 238 (30%) | 0.429 | 44% | 45% | 0.073 |
| 1 | 143 (42%) | 367 (46%) | 6.1% | 5.6% | ||
| 2 | 92 (27%) | 149 (19%) | 23% | 21% | ||
| ≥3 | 21 (6.1%) | 37 (4.7%) | 27% | 28% | ||
| Missing | 143 (29%) | 739 (48%) | 0.00% | 0.00% | ||
| Immunotherapy prescribing rate* | ||||||
| <25% | 219 (45%) | 1063 (69%) | 0.591 | 63% | 64% | 0.036 |
| 25-50% | 194 (40%) | 421 (28%) | 31% | 30% | ||
| 50-75% | 52 (11%) | 43 (2.8%) | 4.9% | 4.9% | ||
| >75% | 22 (4.5%) | 3 (0.20%) | 1.3% | 0.98% | ||
| PD-L1 tested | 41 (8.4%) | 93 (6.1%) | 0.090 | 8.9% | 6.2% | 0.099 |
| PL-L1 result, among tested | ||||||
| Positive | 9 (22%) | 16 (17%) | 0.311 | 22% | 16% | 0.285 |
| Negative | 18 (44%) | 55 (59%) | 47% | 61% | ||
| Unknown | 14 (34%) | 22 (24%) | 31% | 22% | ||
| Mean comorbidity score§ | 2.5 | 1.1 | 0.355 | 1.5 | 1.5 | 0.017 |
| Prescription medication use at treatment initiation± | ||||||
| Any opioids | 133 (27%) | 308 (20%) | 0.169 | 72% | 71% | 0.034 |
| Any steroids | 39 (8.0%) | 210 (14%) | 0.184 | 44% | 45% | 0.016 |
SMD = standardized mean difference; BMI = body mass index; ECOG = Eastern Cooperative Oncology Group; PD-L1 = programmed death-ligand 1
Proportions computed among non-missing patients.
Immunotherapy prescribing rate at the clinic where patients received treatment
Elixhauser et al18
Defined as 60 days prior to 7 days after starting therapy
Unweighted baseline characteristics were generally similar between treatment groups, with two exceptions: immunotherapy initiators had a higher ECOG performance status (ECOG ≥2: 33% versus 24%) and Elixhauser comorbidity score (>5, 14% versus 5.8%) (Table 1). All weighted baseline characteristics included in the propensity score model were well-balanced between treatment groups (Table 1, Supplemental Figure 2). The median time from diagnosis to first-line treatment initiation was 31 days for patients receiving immunotherapy and 35 days for patients receiving carboplatin-based chemotherapy.
Overall Survival
Median follow-up time was 7.2 months (interquartile range [IQR] 3.2 to 14 months), defined as the time from the start of treatment to the earliest of death, data extraction, or last EHR activity. The follow-up time among individuals who remained alive was 11 months (IQR 5 to 22 months). During follow-up, there were 1,219 deaths (n=939 among carboplatin-chemotherapy initiators and n=280 among immunotherapy initiators). IPTW-adjusted Kaplan Meier curves are displayed in Figure 1. Adjusted and unadjusted Kaplan Meier curves were similar (Supplemental Figure 3). The median overall survival was 9 months in the immunotherapy group and 11 months in the carboplatin-based chemotherapy group. Relative to the carboplatin-based chemotherapy group, the estimated overall survival rate in the immunotherapy group was lower at 12 months (40% [95% CI 34-45%] versus 46% [95% CI 43-49%], p=0.05) but higher at 36 months (28% [95% CI 22-35%] versus 13% [95% CI 11-16%], p<0.001) (Table 2). In the first 12 months, treatment with immunotherapy was associated with an increased hazard of death compared to chemotherapy (HR 1.37, 95% CI 1.15-1.62, p<0.001). Among patients who survived one year after initiation of treatment, subsequent survival was improved for immunotherapy compared to chemotherapy (HR 0.50, 95% CI 0.30-0.85, p=0.01). The difference in the immunotherapy effect before and after 12 months was statistically significant (p<0.001). The E-values (relative risk) for the point estimates for death were 1.79 (≤12 months) and 2.59 (>12 months).
Figure 1:
IPTW-adjusted Kaplan-Meier estimates for overall survival
Table 2:
Inverse probability of treatment weighting (IPTW)-adjusted survival outcomes
| First-line immunotherapy (N=487) |
First-line carboplatin-based chemotherapy (N=1,530) |
|
|---|---|---|
| Overall survival (OS)* | Estimate, 95% CI | Estimate, 95% CI |
| Median OS | 9 months | 11 months |
| 12-month OS | 40% (34 – 45%) | 46% (43 – 49%) |
| 36-month OS | 28% (22 – 35%) | 13% (11 – 16%) |
| Hazard ratio ≤12 months | 1.37 (1.15 – 1.62) | 1.00 (reference) |
| Hazard ratio >12 months | 0.50 (0.30 – 0.85) | 1.00 (reference) |
| Second-line therapy-free survival (TFS)§ |
Estimate, 95% CI | Estimate, 95% CI |
| Median TFS | 6 months | 7 months |
| 12-month TFS | 26% (21 – 31%) | 24% (21 – 26%) |
| 36-month TFS | 16% (11 – 22%) | 5.9% (4.5 – 7.4%) |
| Hazard ratio ≤12 months | 1.18 (1.03 – 1.36) | 1.00 (reference) |
| Hazard ratio >12 months | 0.38 (0.20 – 0.71) | 1.00 (reference) |
Defined as the time from the start of first-line therapy to the date of death.
Defined as the time from the start of first-line therapy to the earliest of start date of second-line therapy or death.
Second-line therapy-free survival
There were 818 patients (41%) who received second line therapy. For immunotherapy initiators, 22% of patients received second-line treatment whereas 47% of carboplatin-based chemotherapy initiators received second-line treatment. In the immunotherapy group, 39% (41/106) received a platinum-based regimen as second-line therapy; in the carboplatin-based chemotherapy group, 39% (279/712) received immunotherapy as second-line therapy (Table 3). The follow-up time among patients who remained alive and did not receive second-line treatment was 8 months (IQR 4 to 17 months). During follow-up, there were 1,643 deaths or progression to second-line treatment (n=1,304 among carboplatin-chemotherapy initiators and n=339 among immunotherapy initiators). At 12 months, the estimated second-line therapy-free survival in the immunotherapy group was similar to the carboplatin-based chemotherapy group (26% [95% CI 21-31%] versus 24% [95% CI 43-49%], p=0.5). At 36 months, second-line therapy-free survival was higher in the immunotherapy group (28% [95% CI 22-35%] versus 13% [95% CI 11-16%], p<0.001). Similar to observed associations with overall survival, treatment with immunotherapy was associated with decreased second-line therapy-free survival in the first 12-months (HR 1.18, 95% CI 1.03 to 1.36, p=0.02) and increased second-line therapy-free survival beyond 12 months (HR 0.38, 95% CI 0.20 to 0.71, p=0.003), compared to chemotherapy (Table 2). The E-values for the point estimates for second-line therapy-free survival were 1.50 (≤12 months) and 3.33 (>12 months).
Table 3:
Second-line treatment
| First-line immunotherapy N=487 |
First-line carboplatin-based chemotherapy N=1,530 |
|
|---|---|---|
| Second-line treatment received | ||
| Yes | 106 (22%) | 712 (47%) |
| No | 381 (78%) | 818 (53%) |
| Second-line regimen, among patients receiving second-line therapy | ||
| Immunotherapy | 30 (28%) | 279 (39%) |
| Carboplatin-based chemotherapy | 31 (29%) | 153 (21%) |
| Cisplatin-based chemotherapy | 10 (9.4%) | 37 (5.2%) |
| Other non-NCCN guideline or non-evidence-based therapy* | 48 (45%) | 259 (36%) |
NCCN = National Comprehensive Cancer Network
Therapies include: 1) immunotherapy and carboplatin-based chemotherapy, 2) immunotherapy and non-carboplatin/non-cisplatin-based chemotherapy, 3) immunotherapy and non-chemotherapy, 4) carboplatin and cisplatin-based chemotherapy, 5) carboplatin-based chemotherapy and non-chemotherapy, 6) cisplatin-based chemotherapy and non-chemotherapy, 7) single-agent non-carboplatin/non-cisplatin-based chemotherapy, 8) non-carboplatin/non-cisplatin-based chemotherapy and non-chemotherapy.
Exploratory subgroup analysis
At 6 months, survival was highest for PD-L1 positive patients treated with immunotherapy and lowest for PD-L1 negative patients treated with immunotherapy, relative to patients treated with chemotherapy (Figure 2).
Figure 2:
Kaplan-Meier estimates for overall survival stratified by PD-L1
Discussion
For mUC patients who are ineligible for standard cisplatin-based chemotherapy, no prior studies have directly compared first-line immunotherapy with carboplatin-based chemotherapy. Most of the evidence comes from cross-trial comparisons of response rates rather than comparisons with real-world overall survival. In the absence of comparative data, providers and patients face a difficult choice regarding first-line treatment selection in clinical practice. In this study, we demonstrate that patients treated with immunotherapy had a 37% increase in the hazard of death in the first 12 months after initiation of therapy, but among those who survive one year, there was a 50% lower hazard of death beyond 12 months after initiation of therapy. These results suggest that clinicians and patients should carefully balance the short-term benefit of chemotherapy against the long-term benefit of immunotherapy.
Our observation of decreased short-term survival with immunotherapy is consistent with the preliminary results from data monitoring committees’ early review of two ongoing first-line immunotherapy trials, which showed decreased survival in patients treated with immunotherapy monotherapy relative to the chemotherapy arms. Although exploratory in nature, our PD-L1 stratified analysis also suggests that PD-L1 negative patients have inferior survival with immunotherapy relative to chemotherapy, supporting the EMA and FDA label revision restricting immunotherapy use to mUC patients whose tumors are PD-L1 positive (approximately 30% of all tumors). Future studies should compare patient outcomes associated with a biomarker-guided versus un-guided treatment strategy. While more granular data from the ongoing phase III trials are pending, our results provide critical insight into this label change, suggesting that the risk-benefit profile with immunotherapy is not favorable for all mUC patients. Early decreased survival with immunotherapy relative to chemotherapy may reflect the subset of patients who do not respond to immunotherapy (65% in KEYNOTE-052) or, less commonly, who exhibit a phenomenon of hyper-progression.5,25,26 Therefore, some populations (e.g., symptomatic or high-volume disease) may instead benefit from chemotherapy as initial therapy. However, our data also suggest a long-term benefit of immunotherapy. The long-term benefit may not have been captured in early review of the phase III trials upon which the EMA and FDA’s label restriction was based. Our findings of improved short-term survival with carboplatin-based chemotherapy but superior long-term survival with immunotherapy provide a rationale for considering first-line combination chemotherapy and immunotherapy in an effort to achieve maximal survival for all patients. This is currently being explored in the ongoing trials KEYNOTE-361 and IMVigor-130.
Our reported 40% 12-month OS rate for first-line immunotherapy is modestly lower than that observed in the two phase II trials that led to accelerated approval for immunotherapy, which showed 12-month OS rates of 48% (pembrolizumab) and 57% (atezolizumab).6,27 The higher 12-month OS rate in these studies is likely reflective of the narrow eligibility criteria of the clinical trials, which often exclude patients with multiple comorbid illnesses and the elderly.28 Notably, the long survival tail for patients receiving immunotherapy in our study mirrors findings across disease groups for patients treated with checkpoint inhibitors. In contrast, few patients receiving carboplatin-based chemotherapy survived beyond three years in our study, similar to the results from clinical trials of carboplatin-containing chemotherapy.3
This study had several unique strengths. The large sample size of over 2,000 patients and recency of data allowed us to study long-term effectiveness of immunotherapy as first-line therapy for mUC, a relatively new indication compared with the historical standard of carboplatin-based chemotherapy. In the absence of randomized data, our observations have real-world treatment implications. The analyses were conducted with IPTW models to account for confounding by factors associated with treatment selection and the outcome of interest, an approach with improved power relative to propensity score matching and improved confounding control relative to propensity score adjustment.29
There are several limitations to this study. First, despite including a large range of covariates in our propensity score models, there is risk of residual confounding. Although advanced propensity score-based methods are efficient for reducing bias from imbalance in observed confounders, such methodology does not address unmeasured confounders. For example, we were unable to assess the presence of visceral metastases, which is associated with poor outcome in mUC patients, as data on metastatic sites of disease were unavailable in the database. However, for the outcome of overall survival, the sensitivity analysis using the E-value methods suggest that our observed associations could only be explained by an unmeasured confounder that was associated with both receipt of immunotherapy and risk of death by a risk ratio of more than 1.79 (≤12 months) or 2.59 (>12 months) beyond that of the covariates measured in our study. Furthermore, residual confounding would only be expected to explain the observed treatment effects if the unmeasured confounder was associated with receipt of immunotherapy and with both increased early mortality and decreased late mortality. Because the direction of the association reverses, an unmeasured confounder would either need to be time-varying, with a distribution within treatment groups that changes part way through follow-up, or it would need to have a time-dependent effect that switched direction part way through follow-up. Therefore, while possible, it is unlikely that such an unmeasured confounder exists. Additionally, ECOG performance status, an important predictor of outcomes in cancer patients, required imputation for a large percentage of patients. Multiple imputation has been shown to be effective in EHR research under many missingness patterns as long as the fraction of missing information is not too large.30 Across ten imputed data sets, we observed minimal variation in parameter estimates, suggesting that information lost due to missing data was small (Supplemental Figure 4). Further, we relied on ICD-9/10 codes to identify comorbidity which may be imperfect, as comorbidity may not be documented in an oncology-specific electronic health record unless it affects treatment selection. Second, because PD-L1 testing was not mandated until June of 2018, analyses stratified by PD-L1 status had reduced statistical power—over 90% of patients did not have PD-L1 testing performed. Third, within the confines of the Flatiron Health database, specific criteria were not used to define cisplatin-ineligibility. Rather, patients were assumed cisplatin-ineligible if they received carboplatin or immunotherapy in the first line, consistent with commonly accepted treatment guidelines for cisplatin-ineligible patients. However, only 5.7% of patients received cisplatin in the second line, suggesting that the majority of patients were likely ineligible for front-line cisplatin. Fourth, treatment-related toxicity is not yet available in the Flatiron Health database. Indeed, optimal first-line treatment selection involves consideration of both efficacy and safety. For example, high-grade toxicity, including death, was reported in 15% of cisplatin-ineligible mUC patients treated with carboplatin-based chemotherapy in a prospective clinical trial.3 Although immunotherapy is often considered a well-tolerated option, these therapies are also associated with severe and fatal immune-mediated adverse events.31
Conclusions
In summary, this observational cohort study demonstrated inferior short-term but superior long-term survival with first-line immunotherapy relative to carboplatin-based chemotherapy among patients with metastatic urothelial cancer treated in routine clinical practice. We cannot exclude the possibility that an unmeasured confounder contributed to this association. Choosing between these options will require the identification of patient subgroups that may derive benefit or harm from first-line immunotherapy or chemotherapy. Likewise, understanding patient preference for short- versus long-term benefit with systemic therapy is important, particularly in the setting of metastatic disease. Until the currently pending trial results become available, these results provide important information to facilitate decision-making between physicians and patients.
Supplementary Material
Supplemental Figure 1: CONSORT diagram
Supplemental Figure 2: Distribution of propensity scores
Supplemental Figure 3: IPTW-adjusted and unadjusted Kaplan-Meier estimates for overall survival
Supplemental Figure 4: Ten imputations of IPTW-adjusted Kaplan-Meier estimates for overall survival
Acknowledgments
Funding: This study was supported by the National Institutes of Health under the following award numbers R21CA227613 (to RAH) and K23-CA187185 (to RM).
Footnotes
Disclosures: RM has served as a consultant for Roche/Genentech. RAH has received research funding from Humana. NJM, AC, and BA report employment at Flatiron Health, Inc., which is an independent subsidiary of the Roche Group. JC reports part-time employment at Elekta, Inc.
References
- 1.von der Maase H, Hansen SW, Roberts JT, et al. Gemcitabine and cisplatin versus methotrexate, vinblastine, doxorubicin, and cisplatin in advanced or metastatic bladder cancer: Results of a large, randomized, multinational, multicenter, phase III study. Journal of Clinical Oncology. 2000;18(17):3068–3077. [DOI] [PubMed] [Google Scholar]
- 2.Galsky MD, Hahn NM, Rosenberg J, et al. Treatment of patients with metastatic urothelial cancer “unfit” for cisplatin-based chemotherapy. Journal of Clinical Oncology. 2011;29(17):2432–2438. [DOI] [PubMed] [Google Scholar]
- 3.De Santis M, Bellmunt J, Mead G, et al. Randomized phase II/III trial assessing gemcitabine/carboplatin and methotrexate/carboplatin/vinblastine in patients with advanced urothelial cancer who are unfit for cisplatin-based chemotherapy: EORTC study 30986. Journal of Clinical Oncology. 2012;30(2):191–199. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Necchi A, Pond GR, Raggi D, et al. Efficacy and safety of gemcitabine plus either taxane or carboplatin in the first-line setting of metastatic urothelial carcinoma: A systematic review and meta-analysis. Clinical Genitourinary Cancer. 2017;15(1):23–30. [DOI] [PubMed] [Google Scholar]
- 5.Balar AV, Castellano D, O’Donnell PH, et al. First-line pembrolizumab in cisplatin-ineligible patients with locally advanced and unresectable or metastatic urothelial cancer (KEYNOTE-052): A multicentre, single-arm, phase 2 study. The Lancet Oncology. 2017;18(11):1483–1492. [DOI] [PubMed] [Google Scholar]
- 6.Galsky MD, Rosenberg JE, Powles T, et al. Atezolizumab as first-line therapy in cisplatin-ineligible patients with locally advanced and metastatic urothelial carcinoma: A single-arm, multicentre, phase 2 trial. Lancet. 2017;389(10064):67–76. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.European Medicines Agency. EMA restricts use of keytruda and tecentriq in bladder cancer. https://www.ema.europa.eu/en/news/ema-restricts-use-keytruda-tecentriq-bladder-cancer. Updated 2018. Accessed 4/23/, 2019.
- 8.Food and Drug Administration. Keytruda (pembrolizumab) or tecentriq (atezolizumab): FDA alerts health care professionals and investigators: FDA statement - decreased survival in some patients in clinical trials associated with monotherapy. https://www.fda.gov/Safety/MedWatch/SafetyInformation/SafetyAlertsforHumanMedicalProducts/ucm608253.htm. Updated 2018. Accessed February 2, 2019.
- 9.Food and Drug Administration. FDA limits the use of tecentriq and keytruda for some urothelial cancer patients. https://www.fda.gov/Drugs/InformationOnDrugs/ApprovedDrugs/ucm612484.htm. Updated 2018. Accessed February 2, 2019.
- 10.Jaffe EM, Dang CV, Agus DB, et al. Future cancer research priorities in the USA: A lancet oncology commission. Lancet Oncology. 2017;18(11):e706. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 11.Wang SV, Schneeweiss S, Berger ML. Reporting to improve reproducibility and facilitate validity assessment for healthcare database studies v1.0. Value Health. 2017;8:1009–1022. [DOI] [PubMed] [Google Scholar]
- 12.Abernethy AP, Gippetti J, Parulkar R, Revol C. Use of electronic health record data for quality reporting. Journal of Oncology Practice. 2017;3(8):530–534. [DOI] [PubMed] [Google Scholar]
- 13.Griffith SD, Tucker M, Bowser B, et al. Generating real-world tumor burden endpoints from electronic health record data: Comparison of RECIST, radiology-anchored, and ClinicianAnchored approaches for abstracting real-world progression in non-small cell lung cancer. Advances in Therapy. 2019;Published Online. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Presley CJ, Tang D, Soulos PR, et al. Association of broad-based genomic sequencing& with survival among patients with advanced Non–Small cell lung cancer in the community oncology setting. JAMA. 2018;320(5):469–477. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Singal G, Miller PG, Agarwala V, et al. Association of patient characteristics and tumor genomics with clinical outcomes among patients with non-small cell lung cancer using a clinicogenomic database. JAMA. 2019;321(14):1391–1399. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 16.Mazzone E, Preisser F, Nazzani S, et al. More extensive lymph node dissection improves survival benefit of radical cystectomy in metastatic urothelial carcinoma of the bladder. Clinical Genitourinary Cancer. 2018;18. doi: 10.1016/j.clgc.2018.11.003. [DOI] [PubMed] [Google Scholar]
- 17.Curtis MD, Griffith SD, Tucker M, et al. Development and validation of a high-quality composite real-world mortality endpoint. Health services research. 2018;53(6):4460–4476. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.van Walraven C, Austin PC, Jennings A, Quan H, Forster AJ. A modification of the elixhauser comorbidity measures into a point system for hospital death using administrative data. Medical Care. 2009;47(6):626–633. https://www.jstor.org/stable/40221931. [DOI] [PubMed] [Google Scholar]
- 19.Sterne JA, White IR, Carlin JB, et al. Multiple imputation for missing data in epidemiological and clinical research: Potential and pitfalls. BMJ. 2009;338:b2393. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Rubin DB. Multiple imputation for nonresponse in surveys. New York: John Wiley and Sons; 2004. [Google Scholar]
- 21.Marshall A, Altman DG, Holder RL, Royston P. Combining estimates of interest in prognostic modelling studies after multiple imputation: Current practice and guidelines. BMC Medical Research Methodology. 2009;9(57):Published online. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Austin PC. An introduction to propensity score methods for reducing the effects of confounding in observational studies. Multivariate Behavioral Research. 2011;46(3):399–424. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Pirracchio R, Petersen ML, van der Laan M. Improving propensity score estimators’ robustness to model misspecification using super learner. American Journal of Epidemiology. 2015;181(2):108–119. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Peng D, VanderWeele TJ. Sensitivity analysis without assumptions. Epidemiology. 2016;27(3):368–377. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Soria F, Beleni AI, D’Andrea D, et al. Pseudoprogression and hyperprogression during immune checkpoint inhibitor therapy for urothelial and kidney cancer. World Journal of Urology. 2018;36(11):1703–1709. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Champiat S, Dercle L, Ammari S, et al. Hyperprogressive disease is a new pattern of progression in cancer patients treated by anti-PD-1/PD-L1. Clinical Cancer Research. 2017;23(8):1920–1928. [DOI] [PubMed] [Google Scholar]
- 27.Vuky J, Balar AV, Castellano DE, et al. Updated efficacy and safety of KEYNOTE-052: A single-arm phase 2 study investigating first-line pembrolizumab (pembro) in cisplatin-ineligible advanced urothelial cancer (UC). Journal of Clinical Oncology. 2018;36(15):(suppl; abstr 4524). [Google Scholar]
- 28.Lewis JH, Kilgore ML, Goldman DP, et al. Participation of patients 65 years of age or older in cancer clinical trials. Journal of Clinical Oncology. 2003;21(7):1383–1389. [DOI] [PubMed] [Google Scholar]
- 29.Austin PC. The performance of different propensity score methods for estimating marginal hazard ratios. Statistics in Medicine. 2013;32(16):2837–2849. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Wells BJ, Chagin KM, Nowacki AS, Kattan MW. Strategies for handling missing data in electronic health record derived data. EGEMS (Wash DC). 2013;1(3):1035. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 31.Wang D, Salem JE, Cohen JV, et al. Fatal toxic effects associated with immune checkpoint inhibitors: A systematic review and meta-analysis. JAMA Oncology. 2018;4(12):1721–1728. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Supplemental Figure 1: CONSORT diagram
Supplemental Figure 2: Distribution of propensity scores
Supplemental Figure 3: IPTW-adjusted and unadjusted Kaplan-Meier estimates for overall survival
Supplemental Figure 4: Ten imputations of IPTW-adjusted Kaplan-Meier estimates for overall survival