Incomplete toxicity reporting and use of toxicity-minimizing language in phase III oncology trials

Abstract

Purpose:

To determine the completeness of toxicity reporting (CTR), and the use of subjective toxicity-minimizing language (TML) among phase III oncology trials.

Methods:

Two-arm superiority-design phase III oncology trials published from 2002 and 2020 were reviewed for toxicity data. Complete toxicity reporting, defined as reporting total adverse events (AEs), total serious AEs, total deaths, and study therapy discontinuations due to toxicity; guideline concordance, on the basis of guidelines published in the BMJ (defined as reporting total serious AEs, total deaths, and study therapy discontinuations due to toxicity); and TML, defined as a set of terms that subjectively downplay the harm of therapies.

Results:

A total of 407 trials enrolling 322,645 patients were included. Most (51%, n=207) reported serious AEs, 88% (n=358) reported total deaths, and 84% (n=340) reported study therapy discontinuation due to toxicity. Although 55% of trials (n=223) reported total AEs, only 32% (n=131; 95% CrI 28% to 37%) fit the criteria for CTR. CTR was more common in trials with industry sponsorship (37%) than with cooperative group sponsorship (4%). All 131 trials where CTR was observed were industry sponsored and only 3% (4/131) were cooperative group sponsored trials. Similarly, TML was used in 46% of trials (n=186; 95% CrI, 41% to 51%), with no trial-related factors (including sponsorship source) associated with the odds of TML use.

Conclusion:

Toxicity in phase III oncology clinical trials is often incompletely reported and is frequently minimized in its interpretation by trialists. Industry sponsored trials more comprehensively report toxicity than do cooperative group sponsored trials. CTR may improve patients’ and oncologists’ understanding of new treatments; thus, a more standardized approach to reporting toxicity data is needed.

Introduction

Toxicity outcomes are essential in the interpretation of clinical trial results.¹ The incidence, categorization, grade, severity, duration, and resolution of toxicities, as well as study therapy discontinuation due to toxicity, are critical information for both providers and patients when considering therapeutic approaches.² Often, there are tradeoffs between efficacy and toxicity^3–5 that influence clinical decision-making.

Clinicians’ and patients’ ability to draw clinically meaningful conclusions from clinical trials is dependent on data reporting transparency and completeness.⁶ Prior analyses of trial subsets suggest inconsistent toxicity reporting among disease site–specific oncologic studies, and the widespread use of toxicity-minimizing language (TML), which subjectively minimizes adverse effects (e.g., “manageable,” “acceptable”).^7–9 Prior efforts, including a guideline from 2016,¹⁰ have endeavored to improve the standardization and completeness of toxicity reporting. This 2016 guideline suggests that “best practice” reporting includes information on deaths, serious adverse events, and AEs that led to discontinuation, which constitute the most “clinically relevant” adverse events. However, the current scope and extent of toxicity reporting across late-phase oncology trials and the use of TML, which are directly translated to clinical practice and used for regulatory decisions, remain unknown. Thus, the purpose of this study was to determine the state of toxicity reporting and interpretation among contemporary phase III oncology trials.

METHODS

In this cross-sectional analysis, a search of ClinicalTrials.gov was conducted to identify oncologic clinical trials using the following search parameters: other terms: “cancer”; study type: “All Studies”; status: excluded “Not yet recruiting”; phase: phase 3; and study results: “With Results.” The resulting 1239 trials yielded from this search were then screened to include only phase III randomized two-arm clinical trials in oncology published between 2002 and 2020, with superiority-design and time-to-event primary endpoints. Trials were only included in this study if the primary endpoint results were published in manuscript form. Two individuals (AMM and AHP) reviewed trials, and adjudication was performed by ADS and EBL for ambiguous trials.

Institutional review board approval was not required because all data were publicly available. This study was consistent with STROBE reporting standards for cross-sectional studies.

All manuscripts, including any supplemental materials, were reviewed for toxicity data. A previous publication in 2016¹⁰ recommended the following reporting guidelines: reporting deaths, total serious adverse events (SAEs), adverse events that led to study discontinuation, and specifying AEs of interest; reporting timing, frequency, and duration of AEs; where appropriate, using statistical analysis for clinically-relevant AEs; avoiding overly general descriptions of adverse events; and discussing AE findings in the context of other published evidence. On the basis of these published toxicity reporting guidelines,¹⁰ we modified the definition of guideline-concordance as reporting SAEs, total deaths, and study therapy discontinuation due to toxicity. We also recorded reporting of total adverse events (TAE).

Notably, we did not require that studies report all adverse events by grade, length of time, all side effects, or other data associated with discontinuation of therapy, outside of the total number of patients discontinuing specifically due to toxicity, to be considered for these aforementioned categories. Additionally, we modified the Lineberry et al suggestion to report AEs of interest and instead report TAEs in order to capture reporting of low-grade events, which may still be clinically relevant. Accordingly, because of the frequent clinical significance of chronic, lower-grade toxicities to patients,^11–13 complete toxicity reporting (CTR) was defined as reporting that was guideline-concordant, as described above, and included TAEs. Manuscripts were manually reviewed for TML, defined as in a prior study as verbiage that “could imply downplaying of harms,”⁸ including the terms “acceptable,” “manageable,” “feasible,” “favorable toxicity profile,” “tolerable or well tolerated,” and “safe.” Publication year referred to the year when the primary endpoint results were published by the trial according to PubMed. Disease site was defined as the site of origin of the primary tumor. The definition of a professional medical writer was taken from a previous publication and referred to writers who provide professional assistance in writing the trial manuscripts¹⁴. Cooperative groups were defined as nonprofit organizations that typically receive government funding. Industry was defined as for-profit companies, including pharmaceutical and biotechnological companies; industry sponsorship and cooperative group sponsorship were not mutually exclusive. Treatment type was categorized according to previous definitions (e.g., cytotoxic chemotherapy, targeted therapy, or immunotherapy)¹⁵. Trial outcomes for the primary endpoint (i.e., positive or negative) were defined by whether the primary endpoint met the prespecified statistical definition for significance, and FDA approval of the drug used in the experimental arm was determined by a web search according to previous methods¹⁶.

Logistic regression was used to assess associations between trial-level covariates and study outcomes, including toxicity reporting and TML. Multivariable logistic regressions adjusting for confounding variables computed adjusted odds ratios (aORs) for each trial-level covariate. Confounders specific to each trial-level covariate were determined by mapping casual pathways on a directed acyclic graph using DAGitty (Supplemental Figure 1),¹⁷ as previously reported.^18,19 Each covariate was separately and sequentially selected as the exposure of interest within DAGitty in order to determine its specific confounders. Almost all trials that were not industry sponsored were not guideline concordant; thus, a subgroup analysis was performed exclusively among industry sponsored trials to identify additional trial-level covariates associated with CTR. Chi-square tests were used to assess the correlation between outcome variables. No missing data were encountered. All tests were two-sided, and 95% confidence intervals (95% CIs) were reported with significance defined as P<0.05. A probabilistic sensitivity analysis was performed to generate a 95% credible interval for the main outcomes by taking the random samples from the beta distribution of the findings after 10,000 simulations. Statistical analyses were performed using SAS v9.4 (Cary, NC), and plots were created in Prism v10 (GraphPad, La Jolla, CA).

Patients or the public were not involved in the design, or conduct, or reporting, or dissemination plans of our research.

RESULTS

After screening, 407 trials enrolling 322,645 patients met the inclusion criteria (Figure 1). Trials were published between 2002 and 2020. The majority of trials were industry sponsored (Table 1). Additionally, approximately half of the trials met their primary endpoints, about a third led to US Food and Drug Administration approval of the study drug, and 56% used professional medical writer assistance (Table 1).

Figure 1: Flow Diagram.

Table 1 |.

Trial-level covariates of included phase III clinical trials, overall and grouped by outcome variable

		Guideline concordance			Complete toxicity reporting			Toxicity-minimizing language
	Incidence (frequency)	Yes n (%)	OR	P value	Yes n (%)	OR	P value	Yes n (%)	OR	P value
Total no. of trials	407	181 (44)	-	-	131 (32)	-	-	186 (46)	-	-
Therapy type
Cytotoxic chemotherapy	69 (17)	17 (25)	ref		7 (5)	5.29	<0.01	36 (19)	0.69	0.17
Targeted therapy	238 (58)	129 (54)	3.62	<0.01	89 (68)	7.87	<0.01	102 (55)	1.12	0.77
Immunotherapy	51 (13)	24 (47)	2.72	0.01	24 (18)	2.56	0.07	28 (15)	0.63	0.22
Othera	49 (12)	164 (40)	0.89	0.78	11 (8)	ref		20 (11)	ref
Primary endpoint
Met	220 (54)	103 (57)	0.21	.30	76 (58)	1.27	0.27	97 (52)	0.87	0.48
Not met	187 (46)	78 (43)	Ref		55 (42)	ref		89 (48)	ref
Cooperative group sponsored
Yes	90 (22)	11 (6)	0.12	<0.01	4 (3)	0.07	<0.01	36 (19)	0.74	0.22
No	317 (78)	170 (94)	ref		127 (97)	ref		150 (81)	ref
Industry sponsored
Yes	352(86)	179 (99)	27.42	<0.01	131 (100)	n/a		159 (85)	0.85	0.59
No	55 (14)	2 (1)	ref		0 (0)	n/a		27 (15)	ref
PMW
Yes	227 (56)	131 (72)	3.55	<0.01	102 (78)	4.25	<0.01	108 (58)	1.19	0.39
No	180 (44)	50 (28)	ref		29 (22)			78 (42)	ref
FDA approval
Yes	148 (36)	72 (40)	1.49	0.07	56 (43)	1.30	0.20	66 (35)	0.85	0.59
No	259 (64)	109 (60)	ref		75 (57)	ref		120 (65)	ref
Disease site
Breast	68(17)	30 (17)	2.08	0.04	17 (13)	1.58	0.28	23 (12)	0.71	0.32
Thoracic	86 (21)	49 (27)	3.49	<0.01	39 (30)	3.94	<0.01	38 (20)	1.09	0.79
GI	65 (16)	28 (15)	1.99	0.06	21 (16)	2.27	0.05	36 (19)	1.71	0.12
Hematologic	66 (16)	34 (19)	2.80	<0.01	25 (19)	2.90	<0.01	35 (19)	1.56	0.20
GU	53 (13)	21 (12)	1.73	0.16	17 (13)	2.24	0.06	25 (13)	1.23	0.57
Otherb	69 (17)	19 (11)	ref		12 (9)	ref		29 (16)	ref

Abbreviations: PMW, professional medical writer; GI, gastrointestinal; GU, genitourinary

All regressions are univariate.

^aHormonal or non-systemic therapies

^b“Other” includes central nervous system, skin, endocrine, gynecologic, sarcoma, pediatric, head and neck, and multiple disease sites.

Forty-four percent of trials (181 of 407; 95% CrI, 40% to 49%) reported toxicity in a manner that was concordant with prior guidelines (reporting SAEs, total deaths, and rates of study therapy discontinuation due to toxicity); specifically, 51% (n=207) reported SAEs, 88% (n=358) reported total deaths, and 84% (n=340) reported discontinuation due to toxicity. Over half of the trials (n=238; 58%) included targeted therapy, and only 17% (n=69) included traditional cytotoxic agents. Of the 181 guideline-concordant trials, 179 (98%) were industry sponsored, and 11 (6%) were cooperative group sponsored (Table 1). Of the 55 trials that were not industry sponsored alone, 56% (n=31) reported SAEs, 73% (n=40) reported total deaths, and 53% (n=29) reported study therapy discontinuation due to toxicity.

A univariable regression analysis of trial-level covariates other than industry sponsorship showed that publication year, treatment type, use of a professional medical writer, and certain disease sites were associated with guideline concordance (Table 1). Because of the near-absence of guideline concordance among non-industry sponsored trials, a subgroup analysis of industry sponsored trials was used to evaluate other factors associated with guideline concordance while also adjusting for potential confounding factors. Of these factors, greater odds of guideline concordance were found for trials that used professional medical writer assistance (aOR, 1.92; 95% CI, 1.20 to 3.08; P=0.01) and trials that had been published more recently (aOR, 1.18; 95% CI, 1.09 to 1.28; P<0.01; Supplemental Table 1, Figure 2a). Lower guideline concordance was noted among industry sponsored trials that were co-sponsored by cooperative groups (aOR, 0.29; 95% CI, 0.14 to 0.63; P<0.001; Supplemental Table 1).

Figure 2.

Changes over time in guideline concordant reporting and minimizing language. The shaded regions represent the 95% CI of the slope. (a) Guideline Concordance. (b) Complete Toxicity Reporting. (c) Toxicity-Minimizing Language.

With increasing interest in the role of low-grade toxicity in clinical decision making,^11–13 we defined complete toxicity reporting (CTR) as trials that reported TAEs, SAEs, total deaths, and discontinuation due to toxicity. CTR was observed in 32% of trials (n=131; 95% CrI, 28% to 37%), with 55% of all the trials (n=223) reporting TAEs. Fourteen percent (n=57) of trials reported total SAE but not TAE, thereby not reporting low-grade AEs. Parallel to guideline concordance, all 131 trials that provided CTR were industry sponsored, versus only 3% (n=4/131) of cooperative group sponsored trials. Likewise, CTR was more common in industry sponsored trials (131/352, 37%) than trials with cooperative group sponsorship (4/90, 4%). Similar to guideline concordance, publication year, therapy type, disease site, and use of a professional medical writer were also associated with CTR (Table 1); the same was true even after adjusting for confounders (Table 2, Figure 2b).

Table 2 |.

Multivariable logistic regressions of CTR in industry funded trials

Variable	aORa	95% CI	P value
Factor of Interest
Publication Year	1.25	1.14 to 1.37	<.01
Confounders
Cooperative Group Sponsorship	0.20	0.07 to 0.63	0.01
Use of a PMW	1.62	0.92 to 2.85	0.10
Positive Primary Endpoint	0.85	0.52 to 1.38	0.51
Treatment Type
Cytotoxic Chemotherapy	ref
Targeted Therapy	3.33	1.36 to 8.14	0.01
Immunotherapy	2.66	0.92 to 7.69	0.07
Otherb	4.26	1.27 to 14.30	0.02
Factor of Interest
Treatment Type
Cytotoxic Chemotherapy	ref
Targeted Therapy	3.67	1.57 to 8.65	<.01
Immunotherapy	5.44	2.02 to 14.68	<.01
Otherb	4.54	1.46 to 14.15	0.01
Confounders
Cooperative Group Sponsorship	0.17	0.06 to 0.50	<.01
Factor of Interest
Cooperative Group Sponsorship	0.17	0.06 to 0.48	<.01
Confounders
Disease Site
Breast	1.40	0.59 to 3.34	0.45
Thoracic	3.15	1.43 to 6.94	<.01
Gastrointestinal	1.76	0.75 to 4.11	0.19
Hematologic	2.74	1.18 to 6.39	0.02
Genitourinary	2.04	0.83 to 4.99	0.12
Otherc	ref
Factor of Interest
Use of a PMW	2.25	1.35 to 3.75	<.01
Confounders
Cooperative Group Sponsorship	0.25	0.08 to 0.74	0.01
Factor of Interest
Positive Primary Endpoint	0.97	0.62 to 1.52	0.91
Confounders
Cooperative Group Sponsorship	0.17	0.06 to 0.48	<.01
Factor of Interest
Disease Site
Breast	1.49	0.63 to 3.50	0.36
Thoracic	3.33	1.53 to 7.24	<.01
Gastrointestinal	1.99	0.86 to 4.58	0.11
Hematologic	2.65	1.16 to 6.05	0.02
Genitourinary	2.13	0.88 to 5.12	0.09
Otherc	ref
Confounders: NONE

Abbreviations: PMW, professional medical writer; aOR, adjusted odds ratio

^aORs adjusted for each factor of interest, controlled for the confounders that were determined by the directed acrylic graphs (shown in Supplemental Figure 1a).

^bHormonal or non-systemic therapies

^c“Other” includes central nervous system, skin, endocrine, gynecologic, sarcoma, pediatric, head and neck, and multiple disease sites.

Regarding the interpretation of toxicity data, 46% of trials (186 of 407; 95% CrI, 41% to 51%) used TML. The most commonly used TML terms were “acceptable” (50 [12%]), “manageable” (66 [16%]), and “tolerable” (99 [24%]) (Table 3). TML was most-commonly used in the discussion (180 [44%]) and abstract (78 [19%]) (Table 3). No trial-level factors were associated with lower odds of using TML (Table 1), including after confounder adjustment (Table 4, Figure 2c).

Table 3 |.

Distribution of TML throughout included published phase III randomized controlled oncology trials and frequency of each TML term

Toxicity-minimizing language	Incidence (Frequency %)
Total number of trials using TML	186/407 (46)
Location within the manuscript
Abstract	78 (19)
Introduction	18 (4)
Methods	3 (1)
Results	13 (3)
Discussion/conclusion	180 (44)
Toxicity-minimizing language term
Acceptable	50 (12)
Manageable	66 (16)
Feasible	5 (1)
Favorable toxicity profile	32 (8)
Tolerable or well-tolerated	99 (24)
Safe	24 (6)

^aTML is defined as a set of terms that minimize the potential harm of a treatment.

^bNumbers do not add up to 100%, as minimizing words were counted each time they were found in the manuscript.

Table 4 |.

Multivariable logistic regression of TML

Variable	aORa	95% CI	P value
Factor of Interest
Publication Year	1.04	0.97 to 1.11	0.26
Confounders
Cooperative Group Sponsorship	0.49	0.24 to 1.00	0.05
Industry Sponsorship	0.51	0.22 to 1.18	0.11
Use of a PMW	1.02	0.63 to 1.65	0.95
Positive Primary Endpoint	0.85	0.57 to 1.28	0.43
Treatment Type
Cytotoxic Chemotherapy	ref
Targeted Therapy	0.67	0.38 to 1.19	0.17
Immunotherapy	0.96	0.44 to 2.09	0.92
Otherb	0.62	0.29 to 1.32	0.22
Factor of Interest
Treatment Type
Cytotoxic Chemotherapy	ref
Targeted Therapy	0.68	0.39 to 1.21	0.19
Immunotherapy	1.06	0.50 to 2.24	0.88
Otherb	0.64	0.30 to 1.36	0.25
Confounders
Cooperative Group Sponsorship	0.49	0.25 to 0.98	0.04
Industry Sponsorship	0.5	0.22 to 1.15	0.10
Factor of Interest
Cooperative Group Sponsorship	0.76	0.47 to 1.24	0.28
Confounders
Disease Site
Breast	0.68	0.34 to 1.37	0.28
Thoracic	1.03	0.54 to 1.98	0.92
Gastrointestinal	1.62	0.81 to 3.24	0.17
Hematologic	1.52	0.77 to 3.00	0.23
Genitourinary	1.20	0.58 to 2.47	0.62
Otherc	ref
Factor of Interest
Industry Sponsorship	0.81	0.45 to 1.44	0.47
Confounders
Disease Site
Breast	0.71	0.36 to 1.43	0.34
Thoracic	1.13	0.59 to 2.15	0.72
Gastrointestinal	1.75	0.88 to 2.49	0.11
Hematologic	1.59	0.80 to 3.15	0.18
Genitourinary	1.25	0.61 to 2.57	0.55
Otherc	ref
Factor of Interest
Use of a PMW	1.13	0.72 to 1.79	0.59
Confounders
Cooperative Group Sponsorship	0.51	0.25 to 1.03	0.06
Industry Sponsorship	0.45	0.20 to 1.02	0.06
Factor of Interest
Positive Primary Endpoint	0.85	0.57 to 1.27	0.42
Confounders
Cooperative Group Sponsorship	0.47	0.24 to 0.93	0.03
Industry Sponsorship	0.48	0.21 to 1.07	0.07
Factor of Interest
Disease Site
Breast	0.71	0.35 to 1.41	0.32
Thoracic	1.09	0.58 to 2.07	0.79
Gastrointestinal	1.71	0.86 to 3.39	0.12
Hematologic	1.56	0.79 to 3.07	0.20
Genitourinary	1.23	0.60 to 2.53	0.57
Otherc	ref
Confounders: NONE

Abbreviations: PMW, professional medical writer, aOR: adjusted odds ratio

^aORs adjusted for each factor of interest, controlled for confounders that were determined by the directed acrylic graphs (shown in Supplemental Figure 1b).

^bHormonal or non-systemic therapies

^c“Other” includes central nervous system, skin, endocrine, gynecologic, sarcoma, pediatric, head and neck, and multiple disease sites.

DISCUSSION

In this analysis of toxicity reporting and interpretation across a broad collection of recent phase III oncology trials, we found that toxicity was incompletely reported in most studies, and a near-majority included TML in their interpretations of toxicity findings. Taken together, these data provide a key contemporary understanding of the manner and context in which clinically relevant toxicity data are reported, interpreted, and communicated in the publications of phase III oncology trials. As understanding toxicity represents a fundamental component of conceptualizing the risk-benefit tradeoffs for informing shared-decision making with patients, these findings argue for greater attention to comprehensive toxicity reporting in phase III trials, especially for serious adverse events, and the language with which toxicities are characterized.

These findings, specifically the lack of CTR in the majority of trials, are consistent with previously published research that suggests that AEs are not just incompletely reported but reported on the basis of widely variable criteria. A previous study⁶ examining phase III metastatic solid malignancies for concordance with the CONSORT guidelines for reporting AEs found that 96% reported AEs only if they occurred above a certain threshold or severity (perhaps due to journal requirements), and that guideline concordance only slightly increased over time. Additionally, that study found that 37% of 175 trials did not specify the criteria used to select which AEs were reported. Other studies^9,20 have similarly found inconsistencies in the thresholds for reporting AEs and highly variable criteria for determining which AEs are reported. Three separate review papers found that AE reporting was suboptimal, with median completeness scores indicating adherence to just over half of the guidelines in the checklist.²⁰ Our results regarding toxicity reporting rates complement recent findings that AEs are heterogeneously reported, with trials following different criteria and not adhering well to any set of guidelines.

Excluding TAEs in toxicity reporting necessarily leaves out data regarding grade 1 and 2 AEs. Recent literature suggests that grade 1 and 2 AEs (i.e., AEs of low severity) are often not reported by physicians, who tend to under-evaluate severity and frequency of adverse events, compared with patient evaluations.^11–13 A recent study¹¹ examined the relationship between grade 1 and 2 AEs and self-reported patient bother (agreement with the statement “I am bothered by side effects of treatment” on a Likert scale) and study therapy discontinuation. The results indicated that grade 1 AEs increased the odds of bother by 13% and grade 2 by 35%. Additionally, the frequency of grade 2 AEs increased the odds of study therapy discontinuation by 59%. As these data indicate, while grade 1 and 2 AEs may not require hospitalization, they are significant for patients in that they may be chronic, irreversible conditions that impact their daily lives, and they are often underreported by clinicians. Similar to the discrepancies found between the grades 1 and 2 toxicity reporting of patients and physicians, patient-reported outcomes (PROs) more generally have been found in some studies to be underreported by physicians.^21,22 Therefore, in addition to what is already recommended by current guidelines,¹⁰ we encourage CTR, which includes TAEs along with total SAEs, total deaths, and study therapy discontinuations due to toxicity.

Notably, trial sponsorship was strongly correlated with CTR, with industry sponsored trials being far more likely – and cooperative group sponsored trials far less likely – to completely report toxicity data. Furthermore, the use of medical writers in industry trials was associated with improved toxicity reporting. These observed differences may be due to a multitude of factors relating to the inherent distinctions between industry-sponsored trials and cooperative-group-sponsored trials. First, our findings may be attributable to the fact that cooperative groups broadly-speaking have fewer available resources than the biopharmaceutical industry, including the use of professional medical writers,¹⁴ and may follow reporting guidelines that are less rigorous.^23,24 Industry-sponsored trials, therefore, may be more likely to be written in a stereotyped pattern and format, whereas cooperative group trial reports may be less standardized. It is possible that cooperative-group-sponsored trials lack the homogeneity of writing and publication that industry-sponsored trials have partially due to the many coauthors, committees, and individuals involved in the dissemination of cooperative group trial results. Perhaps these data highlight the need for improved standardization of reporting trial results within the cooperative group system. Another possibility is that industry is more likely to completely report toxicity to fulfill regulatory requirements given considerations such as future applications for regulatory approvals, which cooperative-group-sponsored research tends to be less likely to seek. Regardless of the underlying cause/s, our results highlight the need to improve toxicity reporting quality, particularly by cooperative group trialists and publication teams.

A lack of significant associations between TML use and trial factors supports the widespread nature and practice of TML use. To our knowledge, this study is the largest to focus on TML, with more homogenous trials and a broader scope than previously published studies. We found a higher rate of TML use than did smaller studies,^6,8 suggesting that this issue is widespread and not improving over time. Interpreting toxicities and their relative impact on patients is exceptionally challenging due to the subjectivity and variation between patient experiences. While interpretation of toxicity remains an imperative, ultimately, developing context-specific, patient-centric language tools for interpretation of tolerability, which are ideally well correlated with patient-reported outcomes, represents an important area of future research.

There are several limitations to this study. Our study rated concordance on the basis of guidelines in the BMJ¹⁰ that had been published after most of the trials assessed in our report had been completed. However, these guidelines carry forward a consistent message from CONSORT guideline discussions that date back to 2001–2003.²⁵ Additionally, certain CONSORT/BMJ guideline elements were not readily quantifiable, such as the lack of homogeneity in the duration of time for toxicity data collection. Trials varied in how long after the intervention AE data were collected, and we did not account for these differences, which may allow for more AEs when the time to data collection was longer and fewer AEs when the time to collection was shorter. We also did not account for differences regarding the threshold of frequency for an AE to be reported, whether there was a threshold severity for reporting AEs, or whether AEs were reversible. We also did not assess whether trials reported some but not all adverse events (aside from those categorized as SAEs), or a lack of reporting the criteria used to select which adverse events to report. Prior guidelines for toxicity data reporting that includes the timing, frequency, and duration of AEs among a multitude of other suggestions would render few, if any, trials eligible for our study.¹⁰ Coding of trial-level variables, such as medical writer assistance or trial sponsor, were contingent upon accurate reporting of the trial authors, limiting the interpretation of our correlational analyses. Finally, as our study was focused on toxicity and the interpretation of toxicity, PROs were beyond the scope of the study and were not analyzed.

Collectively, these data indicate that toxicity is incompletely reported and subjectively interpreted with TML across a broad collection of phase III oncology trials. This has a major impact on patients and physicians, who use these data to make treatment decisions. We encourage trialists to make toxicity data more available and stakeholders to be in discussions to improve data reporting. Along with the introduction of new guidelines for toxicity reporting (reporting TAEs, SAEs, total deaths, and study therapy discontinuations), editorial enforcement of these guidelines is necessary to ensure that trialists are presenting toxicity data with as much transparency and consistency as possible.

Context Summary

Key objective:

• Prior smaller analyses of trial subsets suggest inconsistent toxicity reporting and the use of subjective toxicity-minimizing language (TML) in oncology clinical trials, which may hinder data interpretation.

• The current scope and extent of toxicity reporting across late-phase oncology trials and the use of TML remain unknown.

Knowledge generated:

• Our study suggests that toxicity is incompletely reported in the majority of recent phase III oncology trials, and a near-majority include the use of subjective TML in interpreting toxicity findings.

• Industry sponsored trials are far more likely, and cooperative group sponsored trials far less likely, to report complete toxicity data.

Relevance:

• Clear toxicity data reporting and interpretation are necessary for clinicians and patients to draw clinically meaningful conclusions from clinical trials.

Data availability:

Research data are stored in an institutional repository and will be shared upon reasonable request to the corresponding author.