Speed of Adoption of Immune Checkpoint Inhibitors of Programmed Cell Death 1 Protein and Comparison of Patient Ages in Clinical Practice vs Pivotal Clinical Trials

This cohort study assesses the speed with which inhibitors of programmed cell death 1 protein reached eligible patients in practice and compares the ages of patients treated in clinical practice with the ages of those treated in pivotal clinical trials.

Key Points

Question

How quickly have immune checkpoint inhibitors changed clinical practice?

Findings

In this cohort study of 3089 patients, most eligible patients were being treated with inhibitors of programmed cell death 1 protein within 4 months after US Food and Drug Administration approval, with treated patients being significantly older than those studied in the pivotal clinical trials that supported US Food and Drug Administration–approved use.

Meaning

Immune checkpoint inhibitors may have rapidly changed clinical practice among populations of patients who differ substantially from those studied in pivotal clinical trials.

Abstract

Importance

The US Food and Drug Administration (FDA) is increasing its pace of approvals for novel cancer therapeutics, including for immune checkpoint inhibitors of programmed cell death 1 protein (anti–PD-1 agents). However, little is known about how quickly anti–PD-1 agents agents reach eligible patients in practice or whether such patients differ from those studied in clinical trials that lead to FDA approval (pivotal clinical trials).

Objectives

To assess the speed with which anti–PD-1 agents agents reached eligible patients in practice and to compare the ages of patients treated in clinical practice with the ages of those treated in pivotal clinical trials.

Design, Setting, and Participants

This retrospective cohort study, performed from January 1, 2011, through August 31, 2016, included patients from the Flatiron Health Network who were eligible for anti–PD-1 agents treatment of selected cancer types, which included melanoma, non–small cell lung cancer (NSCLC), and renal cell carcinoma (RCC).

Main Outcomes and Measures

Cumulative proportions of eligible patients receiving anti–PD-1 agents treatment and their age distributions.

Results

The study identified 3089 patients who were eligible for anti–PD-1 agents treatment (median age, 66 [interquartile range, 56-75] years for patients with melanoma, 66 [interquartile range, 58-72] years for patients with RCC, and 67 [interquartile range, 59-74] years for patients with NSCLC; 1742 male [56.4%] and 1347 [43.6%] female; 2066 [66.9%] white). Of these patients, 2123 (68.7%) received anti–PD-1 agents treatment, including 439 eligible patients with melanoma (79.1%), 1417 eligible patients with NSCLC (65.6%), and 267 eligible patients with RCC (71.2%). Within 4 months after FDA approval, greater than 60% of eligible patients in each cohort had received anti–PD-1 agents treatment. Overall, similar proportions of older and younger patients received anti–PD-1 agents treatment during the first 9 months after FDA approval. However, there were significant differences in age between clinical trial participants and patients receiving anti–PD-1 agents treatment in clinical practice, with more patients being older than 65 years in clinical practice (range, 327 of 1365 [60.6%] to 46 of 72 [63.9%]) than in pivotal clinical trials (range, 38 of 120 [31.7%] to 223 of 544 [41.0%]; all P < .001).

Conclusions and Relevance

Anti-PD-1 agents rapidly reached patients in clinical practice, and patients treated in clinical practice differed significantly from patients treated in pivotal clinical trials. Future actions are needed to ensure that rapid adoption occurs on the basis of representative trial evidence.

Introduction

Major advancements have taken place in cancer research and drug development during the past 5 years, with the US Food and Drug Administration (FDA) designating more than 45 anticancer agents as breakthrough therapies.¹ Increasing enthusiasm exists for further accelerations in the pace of drug development, as evidenced by the Cancer Moonshot and the 21st Century Cures Act.^2,3 However, such initiatives often aim to accelerate the pace of FDA approval—a process that is already rather brisk for new drugs⁴—rather than the pace at which novel therapeutics are adopted among patients who need treatment. In fact, whereas conventional wisdom suggests it most often takes more than 10 years for innovations to change patient care,^5,6,7,8 little is known about how quickly novel therapeutics are reaching patients with cancer in clinical practice.

One concern about the adoption of novel therapeutics for cancer is that they may be entering the market before data are available to support their use in the general population.^9,10 There is particular concern regarding the adequacy of trial data underlying recent approvals of anticancer agents that often enter the market based on preliminary assessments of risks and benefits.^11,12 Rapid adoption of these agents may harm patients if little is known about the effects of treatment on clinical outcomes.¹³ Slow adoption, however, may harm patients similarly by limiting their access to anticancer agents that are innovative and perhaps efficacious.

In this context, immune checkpoint inhibitors of programmed cell death 1 protein (anti–PD-1 agents agents) provide an ideal opportunity to study the adoption of novel therapeutics after FDA approval. Anti-PD-1 agents boost T-cell–mediated antitumor activity, which in turn can lead to major clinical responses.^{14,15,16,17,18,19,20} Increasing evidence exists to support their use in multiple cancer types, including melanoma, non–small cell lung cancer (NSCLC), and renal cell carcinoma (RCC).^21,22 However, the first FDA approvals for these agents cited preliminary evidence of efficacy, with some approvals granted based on single-arm studies that were neither randomized nor controlled.^23,24

The use of anti–PD-1 agents agents is further complicated—like any new class of agents that first enter to market—by concerns that patients treated in practice might differ in age from patients treated in clinical trials that lead to FDA approvals (pivotal clinical trials).^{25,26,27,28,29} This issue is particularly relevant to checkpoint inhibitors because of age-related changes in immunity that might alter risks and benefits.³⁰ Although most studies^30,31,32 have suggested that checkpoint inhibitors are effective regardless of age, others^33,34 have questioned their efficacy in older adults because the relatively small numbers of older participants in trials preclude a clear understanding of whether age moderates the effects of anti–PD-1 agents treatment. In an era of increasing enthusiasm for checkpoint inhibitors³⁵ but also of concern regarding the generalizability of trial evidence with regard to patient age, it is uncertain whether questions about the ages of trial participants might be influencing the adoption of these paradigm-shifting drugs.

To address these knowledge gaps, we studied the adoption of anti–PD-1 agents agents among patients in real-world practice. Because of uncertainty regarding the use of these agents to treat older patients,³³ we completed 2 additional analyses. First, we assessed the use of each agent across age distributions of patients treated in real-world practice. Second, we compared the age distributions of patients treated in practice with those of patients treated in pivotal clinical trials to assess whether clinical trials matched with the later clinical use of each drug.

Methods

Study Design

We conducted a retrospective cohort study of patients receiving systemic anticancer treatment from January 1, 2011, through August 31, 2016. To ensure adequate follow-up time, we included only patients eligible for treatment with nivolumab or pembrolizumab for previously treated or untreated melanoma, with nivolumab or pembrolizumab for previously treated NSCLC, or with nivolumab for previously treated RCC. We did not assess the rate of adoption of the drug atezolizumab, an inhibitor of programmed cell death 1 ligand 1 (PD-L1), because of limited follow-up time after its approval for urothelial carcinoma on May 18, 2016. The New England Independent Review Board approved the study and waived informed consent because all data were deidentified and collected as part of routine clinical practice.

Data Source

We used electronic health record (EHR) data from the Flatiron Health longitudinal EHR database, which included 233 academic and community oncology practices.³⁶ Although these practices were self-selected, their use of a cloud-based EHR platform allowed for the integration of real-world evidence from a large, geographically diverse cohort of patients with melanoma, NSCLC, or RCC. Furthermore, these patients were similar in age, sex, and race/ethnicity to the US population of patients with melanoma, NSCLC, and RCC according to estimates of disease prevalence in Surveillance, Epidemiology, and End Results data from 2014.³⁷ To create our data sets, we combined structured EHR data with elements from unstructured sources, such as practitioner notes, using technology-based abstraction techniques as described previously.³⁸ All data sets included demographic data, such as age, and clinical data, such as cancer type, disease stage, v-Raf murine sarcoma viral oncogene homologue B1 (BRAF) (OMIM 164757) mutation status, and PD-L1 expressivity.

Cohort Selection

We applied prespecified inclusion and exclusion criteria to construct cohorts (eFigures 1-3 in the Supplement). We included patients 18 years or older who underwent treatment for (1) stage III or IV melanoma, (2) stage I or II melanoma with locoregional or distant recurrence, (3) stage IV NSCLC, (4) stage IV RCC, or (5) stage I, II, or III RCC with metastatic progression. All patients had dates of diagnosis or disease progression on or after January 1, 2011, and at least 2 clinical encounters on or after January 1, 2013. We limited our study to patients eligible for anti–PD-1 agents treatment according to FDA label requirements by cancer type, disease stage, histologic type (for NSCLC), line of therapy (defined as first-, second-, or third-line therapy or greater), and BRAF mutation status (for melanoma) (eTable 1 in the Supplement). We excluded patients with multiple primary cancers and patients with missing data regarding histologic types (for NSCLC) or BRAF mutation status (for melanoma). Because we wanted to assess the adoption of each agent into real-world practice, we excluded lines of therapy that represented clinical trial treatments. We then used medication orders and notes of drug administrations to identify anti–PD-1 agents treatment among eligible patients by age.

Identification of Pivotal Clinical Trials

For each cancer type, we used FDA reviews and labels to identify pivotal clinical trials, which are the primary sources of evidence for efficacy underlying FDA approvals.¹¹ We selected clinical trials that were identified in FDA reviews or labels as the only sources of data regarding efficacy or as the primary sources of data regarding efficacy. For each clinical trial, 2 investigators (J.O. and K.F.) used FDA reviews and labels to separately abstract the sample sizes and age distributions of treatment groups, with differences resolved by consensus. If sample sizes or age distributions were missing from FDA reviews or labels, they were abstracted from publications in peer-reviewed journals. If trial results were not published, they were requested from trial sponsors.

Statistical Analysis

We used descriptive statistics to assess characteristics of patients eligible for anti–PD-1 agents treatment. Patients became eligible for treatment when starting a new line of systemic anticancer therapy (with or without an anti–PD-1 agents agent) after the date of FDA approval in any setting for which they met eligibility criteria (eFigure 4 in the Supplement). Among eligible patients, we used a cumulative uptake measure to identify the total number and proportion of patients receiving each agent each month. All patients becoming eligible or receiving anti–PD-1 agents treatment were retained in the respective numerators or denominators for the remainder of the study to allow for cumulative assessments. Finally, we used χ² tests to compare age distributions of patients treated in real-world cohorts with those of patients treated in pivotal clinical trials and to assess for differences in distributions of anti–PD-1 agents treatment among eligible patients by age, with the level of significance being a 2-sided P < .05. All statistical analyses were performed using R, version 3.3.2 (R Foundation for Statistical Computing).

Results

Baseline Characteristics

Of patients who received systemic anticancer therapy (1601 with melanoma, 12 291 with NSCLC, and 1964 with RCC), 3089 were eligible for anti–PD-1 agents treatment, including 555 with melanoma, 2159 with NSCLC, and 375 with RCC (median age, 66 years; 1742 male [56.4%] and 1347 [43.6%] female; 2066 [66.9%] white) (Table 1 and eFigures 1-3 in the Supplement). Of patients eligible for anti–PD-1 agents treatment, the median ages at diagnosis of advanced stage disease were 66 years for patients with melanoma (interquartile range, 56-75 years) and patients with RCC (interquartile range, 58-72 years) and 67 years for patients with NSCLC (interquartile range, 59-74 years). PD-L1 testing was rare, with PD-L1 expressivity reported for 21 patients with melanoma (3.8%), 274 patients with NSCLC (12.7%), and fewer than 5 patients with RCC (<1.3%) (Table 1).

Table 1. Baseline Characteristics of Patients in Real-world Cohorts^a.

Characteristic	Melanoma (n = 555)	NSCLC (n = 2159)	RCC (n = 375)
Age at metastatic or advanced cancer diagnosis, median (IQR), y	66 (56-75)	67 (59-74)	66 (58-72)
Age at metastatic or advanced cancer diagnosis, y
<49	68 (12.3)	127 (5.9)	29 (7.8)
50-64	189 (34.1)	793 (36.7)	143 (38.1)
≥65	298 (53.7)	1239 (57.4)	203 (54.1)
Sex
Male	375 (67.6)	1104 (51.1)	263 (70.1)
Female	180 (32.4)	1055 (48.9)	112 (29.9)
Unknown	0	0	0
Race/ethnicityb
White	452 (81.4)	1351 (62.6)	263 (70.1)
Black or African American	0	164 (7.6)	26 (6.9)
Asian	0	96 (4.4)	0
Other	33 (5.9)	231 (10.7)	46 (12.3)
Unknown	70 (12.6)	317 (14.7)	40 (10.7)
Stage at initial diagnosisc
I-III	272 (49.0)	0	182 (48.5)
IV	159 (28.6)	2159 (100.0)	193 (51.5)
Unknown or not documented	124 (22.3)	0	0
Smoking status
History of smoking	0	1731 (80.2)	204 (54.4)
No history of smoking	0	406 (18.8)	164 (43.7)
Unknown or not documented	555 (100.0)	22 (1.0)	7 (1.9)
Histologic type
Squamous cell carcinoma	0	529 (24.5)	0
Non–squamous cell carcinoma	0	1630 (75.5)	0
PD-L1 status
Positive	10 (1.8)	122 (5.7)	<5 (<1.3)
Negative or equivocal	11 (2.0)	152 (7.0)	0
Not tested or not known	534 (96.2)	1885 (87.3)	>370 (>98.7)
BRAF status
Mutation positive	177 (31.9)	0	0
Mutation negative	378 (68.1)	0	0
Anti–PD-1 treatments
Yes	439 (79.1)	1417 (65.6)	267 (71.2)
Nivolumab	259 (46.7)	1345 (62.3)	267 (71.2)
Pembrolizumab	153 (27.6)	52 (2.4)	0 (0.0)
Both	27 (4.9)	20 (0.9)	0 (0.0)
No	116 (20.9)	742 (34.4)	108 (28.8)

Abbreviations: IQR, interquartile range; NSCLC, non–small cell lung cancer; PD-L1, programmed death ligand 1; PD-1, programmed death protein 1; RCC, renal cell carcinoma.

^aData are presented as number (percentage) of patients unless otherwise indicated.

^bRace categories with fewer than 5 patients are included in the “other” category to reduce the risk of reidentification.

^cStages were unknown or not documented for fewer than 5 patients with RCC; to reduce the risk of reidentification, they are included in the stage I to III group.

Adoption of Anti–PD-1 Agents

Within 4 months after FDA approval of an anti–PD-1 agents, the proportion of eligible patients who had received anti–PD-1 agents treatment in each cohort surpassed 60% (53 of 70 patients with melanoma [75.7%], 99 of 162 patients with NSCLC [61.1%], and 141 of 209 patients with RCC [67.5%]) (Figure 1). At the end of the study period, the numbers of eligible patients having received anti–PD-1 agents treatment were 439 patients with melanoma (79.1%), 1417 patients with NSCLC (65.6%), and 267 patients with RCC (71.2%).

Figure 1. Cumulative Uptake of Treatment With Anti–PD-1 Agents Among Eligible Patients.

Patients who were eligible for or treated with an anti–PD-1 agents agent remained in the numerator or denominator for the rest of the study period. Because some patients received both nivolumab and pembrolizumab, total monthly percentages may exceed 100%. Vertical dashed lines identify the date of initial US Food and Drug Administration (FDA) approval of each agent. Dates of subsequent FDA approvals are given in eTable 1 in the Supplement. PD-1 indicates programmed cell death 1 protein.

There was rapid uptake of pembrolizumab for melanoma in the first 3 months after FDA approval, with 31 of 44 eligible patients (70.4%) receiving anti–PD-1 agents treatment. During the next 12 months, nivolumab surpassed pembrolizumab as the preferred anti–PD-1 agents agent for melanoma (Figure 1). We found rapid adoption of nivolumab but not pembrolizumab in the treatment of NSCLC, with 72 of 2008 eligible patients (3.6%) receiving pembrolizumab. Overall, 27 patients received both anti–PD-1 agents agents (26 sequentially and 1 concurrently) despite little or no evidence supporting such use.

Pivotal Clinical Trials Supporting FDA Approvals of Anti-PD-1 Agents

We identified characteristics of patients treated in pivotal clinical trials underlying initial FDA approvals of each anti–PD-1 agents agent across cancer types and subtypes (eTable 2 in the Supplement). Single pivotal trials supported each approval except for nivolumab in squamous NSCLC, which was supported by 2 pivotal trials. Four of the 7 pivotal trials reported objective responses as the primary outcome, and all 7 reported median follow-up times of 1 year or less. Treatment groups ranged in size from 61 to 410 patients, and their median ages ranged from 57 to 62 years. A total of 465 (38.0%) of these patients were 65 years or older (range, 38 of 120 [31.7%] to 223 of 544 [41.0%]), and 102 (8.8%) were 75 years or older (range, 34 of 410 [8.3%] to 14 of 120 [11.7%]).

Comparisons of Patients in Real-world Cohorts With Patients in Pivotal Trials

Larger proportions of anti–PD-1 agents–treated patients were 65 years or older (range, 827 of 1365 [60.6%] to 46 of 72 [63.9%]) or 75 years or older (range, 74 of 267 [27.7%] to 105 of 286 [36.7%]) in real-world cohorts than in pivotal trials (ranges, 38 of 120 [31.7%] to 223 of 544 [41.0%] in those ≥65 years old and 34 of 410 [8.3%] to 14 of 120 [11.7%] in those ≥75 years old; pairwise P < .001 for all comparisons) (Table 2). However, with up to 9 months of follow-up for each cohort, the proportional uptake of each agent was similar across distributions of patient age (Table 3). This finding suggests that patient age is unlikely to be associated with consistent differences in the clinical adoption of anti–PD-1 agents agents (Figure 2).

Table 2. Comparison of Age Distributions at Receipt of Anti–PD-1 Treatment Among Patients Treated in Pivotal Trials vs in Real-world Cohorts.

Group	Pivotal Trialsa	Real-world Cohorts	P Value
Nivolumab
Melanoma
Age, y
<65	82 (68.3)	110 (38.5)	<.001
65-74	24 (20.0)	71 (24.8)
≥75	14 (11.7)	105 (36.7)
Total	120 (100)	286 (100)	NA
Age, median (range), y	58 (25-88)	69 (27-98)	NA
NSCLC
Age, y
<65	321 (59.0)	538 (39.4)	<.001
65-74	176 (32.4)	423 (31.0)
≥75	47 (8.6)	404 (29.6)
Total	544 (100)	1365 (100)	NA
Age, median (range), y	62 (37-87)	68 (29-94)	NA
RCC
Age, y
<65	257 (62.7)	105 (39.3)	<.001
65-74	119 (29.0)	88 (33.0)
≥75	34 (8.3)	74 (27.7)
Total	410 (100)	267 (100)	NA
Age, median (range), y	62 (23-88)	67 (40-90)	NA
Pembrolizumab
Melanoma
Age, y
<65	59 (66.3)	69 (38.3)	<.001
65-74	23 (25.8)	45 (25.0)
≥75	7 (7.9)	66 (36.7)
Total	89 (100)	180 (100)	NA
Age, median (range), y	57 (18-88)	68 (32-95)	NA
NSCLCb
Age, y
<65	40 (65.6)	26 (36.1)	<.001
≥65	21 (34.4)	46 (63.9)
Total	61 (100)	72 (100)	NA
Age, median (range), y	60 (NA)	70 (38-89)	NA

Abbreviations: NA, not applicable; NSCLC, non-small cell lung cancer; PD-1, programmed death protein 1; RCC, renal cell carcinoma.

^aAdditional details regarding pivotal trials are given in eTable 2 in the Supplement.

^bAdditional age breakdowns and age ranges beyond 65 years were not available for the subgroup of 61 patients from the study by Garon et al,²³ who were included in the primary analysis of efficacy that supported US Food and Drug Administration approval.

Table 3. Patients Receiving Anti–PD-1 Treatment by Age During the First 9 Months After FDA Approvals by Cancer Type .

Group	No. (%) of Patients		P Value
	Anti–PD-1 Treated	Non–Anti–PD-1 Agents Treated
Nivolumab
Melanoma
Age, y
<65	20 (40.0)	30 (60.0)	.81
65-74	14 (46.7)	16 (53.3)
≥75	15 (45.5)	18 (54.5)
Total	49 (43.4)	64 (56.6)
NSCLC
Age, y
<65	123 (58.3)	88 (41.7)	.51
65-74	112 (55.2)	91 (44.8)
≥75	114 (61.0)	73 (39.0)
Total	349 (58.1)	252 (41.9)
RCC
Age, y
<65	98 (77.8)	28 (22.2)	.04
65-74	78 (63.4)	45 (36.6)
≥75	68 (73.9)	24 (26.1)
Total	244 (71.6)	97 (28.4)
Pembrolizumab
Melanoma
Age, y
<65	32 (64.0)	18 (36.0)	.97
65-74	18 (62.1)	11 (37.9)
≥75	21 (61.8)	13 (38.2)
Total	71 (62.8)	42 (37.2)
NSCLC
Age, y
<65	21 (3.1)	653 (96.9)	.50
65-74	14 (2.7)	512 (97.3)
≥75	20 (3.9)	489 (96.8)
Total	55 (3.2)	1654 (96.8)

Abbreviations: NSCLC, non–small cell lung cancer; PD-1, programmed death protein 1; RCC, renal cell carcinoma.

Figure 2. Proportional Uptake of Treatment With Anti–PD-1 Agents Among Eligible Patients by Age.

Vertical dashed lines identify the date of initial US Food and Drug Administration (FDA) approval of each agent. Dates of subsequent FDA approvals are given in eTable 1 in the Supplement. PD-1 indicates programmed cell death 1 protein.

Discussion

We found rapid adoption of anti–PD-1 agents agents after FDA approvals, including after approvals that were based on limited evidence with regard to the sample sizes and age distributions of trial participants. Such rapid adoption stands in contrast to older estimates that suggest it takes years or even decades for new treatments to be adopted,⁵ including evidence that highly effective treatments, such as tamoxifen for breast cancer, can take more than 10 years to reach most eligible patients.³⁹ Therefore, our findings are both encouraging—because physicians are rapidly responding to approvals of novel treatments and incorporating them into practice—and concerning—because rapid adoption of treatments might occur without an adequate understanding of risks and benefits.⁴⁰ Because our sample included mostly community-based practices, our findings also contradict the notion that early access to treatment is limited to academic centers. Finally, we found significant differences in age between patients treated in practice and those treated in trials, which highlights the need to clarify the risks and benefits of checkpoint inhibitors in general populations of patients.

Several factors might have contributed to rapid adoption, including high disease severity, overall preference for novelty, perceived gains over existing treatments, and promotional activities, which include media reports and direct to consumer advertisements.^9,41 Three of these factors might best explain rapid adoption of anti–PD-1 agents agents. First, anti–PD-1 agents agents offered large gains in efficacy to selected patients in part because these agents led to better survival in trials but also because they at times led to major and durable responses to treatment.⁴² This property is understandably important to patients with advanced cancers, who often have poor prognoses and few options for treatment. Second, adverse effects of checkpoint inhibitors are different from those of traditional cancer treatments. It is possible that differences in adverse effect profiles might have further reinforced the early use of anti–PD-1 agents agents. Third, there have been high levels of spending to market these drugs to practitioners and patients.⁴³ In fact, amid unprecedented levels of spending on direct to consumer advertisements of drugs, both nivolumab and pembrolizumab, which might benefit less than 10% of cancer patients,⁴⁴ have ranked among the most advertised drugs of all types.⁴⁵

Our finding of rapid adoption is notable in part because of wide variation in the strength of trial evidence underlying recent FDA approvals.^11,12 We found similar variation underlying accelerated approvals of anti–PD-1 agents agents (eTable 2 in the Supplement). Accelerated approvals, in particular, have been criticized for their reliance on nonrandomized trials with short follow-up periods.^46,47,48 Such approvals are granted pending confirmation of risks and benefits in postapproval studies, which is not always done.^48,49,50,51 In fact, a recent study⁴⁸ found that 50% of postapproval studies were completed for agents approved between 2009 and 2013, with postapproval studies being similar to preapproval studies in their limited use of randomization. Postapproval studies are particularly relevant given the rapid uptake of nivolumab and pembrolizumab after accelerated approvals because accelerated approvals are associated with higher rates of adverse events.^52,53 One disadvantage of approvals based on limited evidence is that they might alter the balance between risks and benefits of costly new drugs. Still, there are several reasons to support faster approvals. Expedited approvals are associated with faster drug development,⁵⁴ and it has been rare for these programs to support approvals of oncology products that are withdrawn because of concerns about safety or efficacy.⁵⁵ As FDA officials develop more flexible standards for approval, which the 21st Century Cures Act requires them to do,³ it is possible that many patients will receive drugs before much is known about clinical outcomes. It will be critical in this context for FDA officials to clarify core principles, to apply such principles consistently, and to include patients in decisions regarding risks that are tolerable in exchange for more rapid access to drugs.⁴⁰ In this context, further integration of real-world evidence might allow the FDA to better assess the drugs that they approve on the basis of nonrepresentative trial participants.^56,57

Real-world evidence is important in particular because of findings of substantial differences in age between patients receiving anti–PD-1 agents agents in practice and those receiving the agents in pivotal trials.⁵⁷ Although data suggest that outcomes are similar between older and younger patients receiving anti–PD-1 agents agents for melanoma,³¹ there is little evidence to guide anti–PD-1 agents treatment of older patients with NSCLC.^34,58 To improve the evidence that supports the treatment of older patients, the FDA plans to adopt recommendations from the American Society of Clinical Oncology,⁵⁹ which include strengthening the FDA’s authority to regulate trial enrollment, integrating tools into trials to assess age-specific outcomes, and requiring journals to report outcomes by patient age.⁶⁰

One key challenge in oncology is to take advantage of the strengths of EHR platforms to study the effects of novel treatments on clinical end points and patient-reported outcomes. At the same time, there is a need to improve the veracity and utility of EHR data sets, which often are cluttered with inputs to support billing rather than clinical practice. In our study, we have found that novel, integrated sources of real-world evidence can provide important insights about the adoption of drugs. However, given such rapid adoption, it is also critical to leverage real-world evidence to further evaluate how well these drugs work.

Limitations

Our study has several limitations. First, because we studied a sample of mostly community-based practices, our findings might not be generalizable to academic or safety-net practices, and we cannot exclude the possibility that adoption might be faster in practices that use novel EHR platforms. Second, because we assessed anti–PD-1 agents agents in FDA-approved contexts, future studies are needed to assess their use in contexts without FDA approval (off label) and to assess how differences in coverage or costs might underlie differences in adoption, for example, among anti–PD-1 agents agents in NSCLC treatment. Third, because we designed a study to assess adoption, about which little is known, we did not assess clinical outcomes, which can be assessed in future work. Fourth, because this analysis is a snapshot of rapidly changing knowledge, practice patterns may change quickly with regard to the use of anti–PD-1 agents treatment. For example, PD-L1 testing may increase as clinicians become more aware of the use of this test.

Conclusions

In summary, we found rapid adoption of anti–PD-1 agents treatment among patients in practice. This finding indicates that clinical practice can change promptly and substantially when novel cancer therapeutics first enter the market—in opposition to traditional teaching, which holds that clinicians are slow to respond to new evidence. Although most studies have confirmed the effectiveness of anti–PD-1 agents treatment in FDA-approved contexts,^61,62 studies have also found that their safety and efficacy can vary across cancer types and lines of therapy.^63,64,65 This variability is a reminder that postapproval studies are imperative when novel treatments enter the market. Furthermore, reasonable enthusiasm for such treatments should be tempered by an expectation that later studies might not support their broad use.¹⁰ Our second major finding—that patients treated in practice were significantly older than patients treated in trials—reflects gaps in knowledge of outcomes among trial participants who represent real-world patients. Future efforts are needed to ensure that the FDA bases its approvals on more generalizable evidence to support the rapid adoption of drugs.

Supplement.

eTable 1. FDA Approval Language and Approval Dates by Cancer Type and Line of Therapy

eTable 2. Characteristics of Pivotal Trials for Each Cancer Type and Subtype

eFigure 1. Melanoma Cohort Formation

eFigure 2. NSCLC Cohort Formation

eFigure 3. RCC Cohort Formation

eFigure 4. Schema for Dynamic Assessments of Eligibility for Anti-PD1 Treatment