Association of patient-reported pain with survival in bladder cancer: a post-hoc analysis of the iBLAD trial

Introduction

This post-hoc analysis of the iBLAD trial investigates the association of patient-reported symptoms with survival in bladder cancer (BC) patients during and after oncological treatment.

Patient-reported outcomes (PRO) are increasingly being used in clinical and research settings to assess self-reported symptoms during oncological treatment [1,2]. The FDA and EMA recommend the inclusion of PROs in all clinical research studies [3,4]. Furthermore, the active use of PROs compared to standard care for symptom-handling during cancer therapy has been shown to improve QoL and even survival [5,6]. More studies have demonstrated how PROs can be used as a complementary tool to assess, monitor and manage symptoms during oncological treatment [7–10].

Several studies have also shown that patient-reported outcome measures can serve as a prognostic tool [6,9,11–13]. PRO measures associated with decreased survival include fatigue, global health, quality of life, pain, loss of appetite and physical function [6,9,11,13–15].

This study aims to evaluate the prognostic significance of patient self-reported symptom scores using the Patient-Reported Outcomes version of the Common Terminology Criteria for Adverse Events (PRO-CTCAE, henceforth PRO) and secondly to identify symptoms associated with reduced survival in BC.

Materials and methods

The present study is a post-hoc analysis of PRO questionnaire responses in the intervention arm of a randomized clinical trial (NCT03584659) [16,17]. The data presented have not been previously published. Feasibility, pilot study, symptom, and item selection for iBLAD were performed in previous studies [18–20]. The PRO questionnaire covered 16 symptoms (32 items) from the PRO-CTCAE library [21]. Locally advanced disease was defined as patients with T2-TxNx disease, and metastatic disease was defined as patients with M1 disease. No patient had started treatment before randomization.

Treatment was carried out according to Danish national guidelines, which follow the European guidelines for BC [22,23]. Physicians and nurses at the treatment centers had access to PRO responses, which were checked at each patient visit, and further an alert algorithm, guided the patient on how to manage the given symptoms, either by symptom-specific self-management or by contacting the treating department (Figure S1). Standard clinical practice for symptom management was used when contacting the department, following local guidelines without a specific management algorithm. Each PRO question has five possible responses, ranging from no symptoms to high symptom burden (Tables S1 and S2).

Figure 1. Kaplan–Meier curve of baseline pain levels. Log-rank/ Mantel–Cox test is performed, p-value in graph.

Table 1. Showing patient characteristics in the group.

	Characteristics	Details	Follow up time
Age	(range)	69 years (44-87 years)
Stage	Locally advanced (n = 45)	9 events	93 days (14-680)
	Metastatic (n = 68)	41 events	151 days (21-872)
Gender	Female (n = 27)
	Male (n = 68)
Treatment	Car/Gem (n = 18)	1 LA, 17 MD
	Cis/Gem (n = 66)	43 LA, 23 MD
	Pembro (n = 28)	1 LA, 27 MD
	Vinflunin (n = 1)	1 MD

MD: metastatic disease; LA: Locally advanced.

Table 2. Multivariate cox-regression analysis for baseline PRO-data and time-varying PRO-data showing cough and pain adjusted for stage, gender, treatment (not-shown, not significant) and age (not shown, not significant).

		Baseline PRO-data				Time varying PRO-data
Variable	Units	n*	HR	CI.95	p-value	n**	HR	CI.95	p-value
Cough (S)	Low <2	(n = 82)	Ref			n = 1037	Ref
	Medium 2-3	(n = 21)	1.41	[0.59;3.37]	0.4440	n = 365	1.77	[0.89;3.51]	0.1054
	High >3	(n = 10)	3.25	[1.15;9.16]	0.0259	n = 84	1.23	[0.25;6.08]	0.8022
Pain (S)	Low <2	(n = 55)	Ref			n = 808	Ref
	Medium 2-3	(n = 28)	2.39	[0.98;5.81]	0.0553	n = 375	0.70	[0.29;1.67]	0.4237
	High >3	(n = 30)	6.05	[2.33;15.71]	<0.001	n = 303	3.17	[1.61;6.25]	<0.001
Stage	Locally advanced	(n = 45)	Ref				Ref
	Metastatic	(n = 68)	2.07	[0.82;5.26]	0.1255		1.95	[0.91;4.17]	0.0857
Gender	F	(n = 27)	Ref				Ref
	M	(n = 68)	0.86	[0.42;1.77]	0.6832		0.80	[0.38;1.69]	0.5559
C-statistics		(n = 113)		c = 0.7887	SE = 0.0378	(n = 1486)		c = 0.7337	SE = 0.0531

*The number of patients. **The number of answers used in analysis. Highly significant findings marked in bold.

Statistics

We used a Mann–Whitney U test to determine differences in age and follow-up time between male/female and locally advanced/metastatic. In addition, a generalized linear binomial regression model examined the difference between events (deaths) in the locally advanced and metastatic groups.

A symptom selection model was built in three steps. First, a crude Cox regression analysis was performed on all 27 symptoms and corresponding items to identify those symptoms/items that correlated with survival using baseline PRO scores. Those symptoms/items with HR >1 and p-value <0.05 were selected for further multivariate analysis. For symptoms where multiple items (frequency (F), severity (S) and interference with daily activities (I)) were significant, the item with the highest HR and lowest p-value was selected. For significant symptoms from the final multivariate baseline analysis, we applied time-dependent covariates (PRO-symptoms) [24]. C-statistics with concordance rates were performed for all Cox regression models to assess model fit.

Baseline and time-dependent survival analysis models were adjusted for disease stage, age, treatment, and sex. Proportionality was assumed for the final models. In the case of missing data on the date of the event, the patient was censored at the date of the last completed questionnaire. No imputation or extrapolation of missing data was performed.

We used a mixed effects model to control for fixed and random effects of covariates over time, as multiple correlated PRO measures are taken for each patient.

Significant symptoms were divided into three strata, low (symptom score <2), medium (2–3) and high (>3), for visualization with Kaplan–Meier survival plots, which correspond to the visual representation of the symptoms as displayed to the clinician in the electronic symptom reporting system (i.e. green, orange, red).

The primary outcome assessed was the symptom(s) with the greatest impact on survival. A p-value <0.05 was considered significant.

All graphs and statistical analyses were performed in R v. 4.2.0 [25].

The analysis followed the recommendations of the Setting International Standards in Analyzing Patient-Reported Outcomes and Quality of Life Endpoints Data Consortium (SISAQOL) guidelines [26] and the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: guidelines for reporting observational studies [27].

Ethical considerations

The study was approved by the Danish Data Protection Agency (suite nb: RH-2017-348), registered at www.clinicaltrials.gov under NCT03584659, and all patients provided written informed consent prior to study entry and randomization. Data protection followed the Danish application of the European GDPR regulations. We adhered to the Declaration of Helsinki and its amendments [28].

Results

From 22 January 2019 to 19 March 2021, 119 patients were enrolled in the intervention arm from four treatment centers in Denmark. Six patients had missing PRO data, leaving responses from 113 patients for further analysis (Figure S1). 99% of weekly questionnaires were completed within the first 200 days after randomization, and three patients completed questionnaires after 200 days after randomization, beyond the planned 6 cycles (approximately 164 days) (Table S3). Most patients responded to the baseline questionnaires within five days of randomization (Figure S2). 82% of all events (deaths) occurred in the MD group, with OR = 6.07 (95% CI 2.53–14.6).

Figure 2. Forest plot showing multivariate time-dependent survival analysis with HR and (CI95%). P-values are shown in Table 3.

Table 3. Multivariate cox-regression with the interaction of pain and stage of disease, both for baseline and time-varying dependents.

	Baseline PRO-data			Time varying PRO-data
Variable	HR	CI.95	p-value	HR	CI.95	p-value
Pain (S) Low: Metastatic vs Local	1.33	[0.26;6.74]	0.7271	2.13	[0.72;6.34]	0.1728
Pain (S) Medium: Metastatic vs Local)	2.93	[0.49;17.74]	0.2410	1.76	[0.33;9.35]	0.5066
Pain (S) High: Stadium(Meta vs Lokal)	2.29	[0.67;7.84]	0.1869	1.77	[0.40;7.79]	0.4531
Locally advanced: Pain (S) (medium vs low)	1.28	[0.16;10.47]	0.8196	0.81	[0.14;4.62]	0.8128
Locally advanced: Pain (S)(high vs low)	3.96	[0.67;23.55]	0.1299	3.70	[0.66;20.74]	0.1364
Metastatic: Pain (S)(medium vs low)	2.81	[1.01;7.82]	0.0480	0.67	[0.24;1.85]	0.4373
Metastatic: Pain (S) (high vs low)	6.80	[2.33;19.90]	<0.001	3.06	[1.46;6.43]	0.0031
C-statistics	c = 0.7925	SE = 0.0373		c = 0.7282	SE = 0.7282

Adjusted for: Cough (S), treatment, stage of disease, age and gender.

We included age, treatment, stage and sex as covariates in the final survival analysis. Patient characteristics and treatments are shown in Table 1. The preliminary steps for symptom identification are shown in Table S4. Baseline multivariate results are shown in Table 2. Briefly, baseline pain (S) remained significant when stratified as high versus low, with an HR at 6.05 (95% CI 2.33–15.71). For baseline cough (S), the HR was 3.25 (95% CI 1.15−9.16) (Table 3).

In the time-dependent analysis of pain (S) (high vs. low), we found an HR of 3.17 (95% CI 1.61–6.25) and for cough (S) HR of 1.23 (95% CI 0.25−6.08). Pain (high vs. low) was significant in the metastatic setting with an HR of 3.06 (95% CI 1.46–6.43) (Table 3), but not in the locally advanced setting (HR 3.7, 95% CI 0.66–20.74). Interaction analysis is shown in Table S5.

When applying a mixed effects model to longitudinal pain scores, we could not show a significant change in pain scores over time (p = 0.46), although the mean pain scores indicated a decrease over time (Figure S3).

Kaplan-Meier curves (Figure 1) show a decreased survival probability for the group of patients with higher pain scores (log-rank 0.0015) (Figure 2).

Discussion

We show in this study that high levels of self-reported pain is significantly associated with survival in patients with metastatic bladder cancer, highlighting the importance of symptom recognition and management.

One may wonder whether the clinical characteristics and treatment of our population differ from those in the literature, and whether differences in follow-up time between the MD and LA may influence our results. The median follow-up time for patients in the LA group was shorter than in the MD group; this is probably due to censoring at the time of referral to urology after neoadjuvant chemotherapy/downstaging and shorter planned treatment. The shorter follow-up time in the LA group is therefore to be expected and warrants interaction analysis between the stage of disease and symptoms of interest. In the interaction analysis adjusting for stage, the pain was only significantly associated with survival for MD. The shorter follow-up time for the LA group probably does not explain the fewer events, as this group of patients is known to have a better prognosis than the MD group. The gender distribution of bladder cancer in our cohort is similar to the known distribution, with a ratio of approximately three to one [29,30]. Similarly, the distribution of therapy administered for the locally advanced and metastatic stages is recognizable from clinical practice and guidelines at the time of the study [31]. The age of the enrolled patients is comparable to the literature [32–35] and the demographics in Denmark [36,37]. The age difference between LA (median 66 years) and MD (median 73) was significantly different, which may also affect survival in the groups in addition to the known worse prognosis for MD patients [38].

It is known that a survivorship effect can bias results when there is a long time between randomization and baseline questionnaires [26]. Most patients responded to the baseline questionnaire within five days of randomization (Figure S2), so the variation in time between treatment initiation and symptom reporting was relatively small. Therefore, a survivorship effect (i.e. survivorship bias) is unlikely. It is noteworthy that pain did not improve significantly during the study period, as one would imagine that pain management would alleviate symptoms. This argues for an increased focus on pain management and timely evaluation of its effects.

Patient characteristics influence treatment outcome and survival [39–41], so we included key patient characteristics such as age, sex, stage of disease and treatment in the final multivariate Cox model. The choice of treatment was not significantly associated with survival in the multivariate analysis, but the fit of the model was improved (data not shown). Previous studies using baseline QOL as a predictor of survival in bladder cancer patients have made similar adjustments [11,42]. In all models, the pain remains significant as a negative predictor of survival for patients in the metastatic setting. The cough does not remain significant in the time-dependent survival analysis (Table 2), and we do not consider this symptom to be prognostic for survival.

It is known from previous symptom and PRO studies that fatigue, pain and nausea are among the most common symptoms for patients undergoing cancer treatment [43]. However, we did not find an association between self-reported fatigue and survival in any of the models. This could be due to treatment selection at baseline, where patients with high levels of fatigue are not eligible for treatment. One can also imagine that recall bias could potentially affect the validity of our data. However, this bias is likely to be almost negligible in this study, as patients responded to questionnaires on a weekly basis.

Pain may be a surrogate marker for high tumor burden and metastatic lesions, which are associated with poorer prognosis [44]. PS is known to be closely correlated with pain [45,46], and improving pain management may improve PS [42]. We seek to identify patient-reported symptoms that may help to identify patients who may need special attention, regardless of disease stage. Possibly, the addition of pain in a model that also incorporates PS would offer more nuanced prognostic information. Due to design issues this study is hypothesis generating, and cannot determine if pain reflects PS, is independent of PS, or if further associations exist. Our results nonetheless show the clinical significance and importance of recognizing patients with pain. A high level of pain should prompt the clinician to consider discussing the importance of managing this symptom with the patient before and during treatment.

This study is limited by its post-hoc analytical design with unmeasured confounders and the inability to evaluate alternative prognostic parameters, i.e. data on pain medication, radiotherapy for painful lesions, metastatic burden, site of metastasis, PS and site of pain.

Our findings highlight the importance of identifying and managing pain at baseline and during treatment in patients with metastatic bladder cancer. This should be investigated in a prospective trial with an increased focus on pain management, i.e. through a nurse-led outpatient pain clinic linked to oncological treatment. The use of real-world PRO data collected from clinics using PRO in daily practice to identify symptoms associated with decreased survival would be of interest for future studies.

Conclusion

In this post-hoc analysis of weekly PRO data during oncological treatment, we find that PRO data can be valuable in predicting survival and that high pain scores, both at baseline and during treatment, are associated with decreased survival in patients with metastatic bladder cancer.

Author contributions

D.R.S: Writing of first draft, formal analysis, revisions of manuscript, study design. G.A.T: Study design, collection of data, revisions of manuscript. R.B.F: Revisions of manuscript. C.J: Study design, revisions of manuscript. H.P: Study design, funding, revisions of manuscript. All authors have read and approved the final manuscript.

Acknowledgements

We thank the statisticians at the Section of Biostatistics at University of Copenhagen for assisting with the formal analysis.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data that support the findings of this study are available from the corresponding author, D.R.S, upon reasonable request.

Correction Statement

This article has been corrected with minor changes. These changes do not impact the academic content of the article.

Additional information

Funding

The iBLAD study and this post-hoc analysis was funded by The Danish Cancer Society, Dagmar Marshalls Fond, Einar Willumsens Mindelegat, A.P. Møller Lægefonden, Christian Larsen og Dommer Ellen Larsens Legat, Rigshospitalets Fond til støtte for onkologiske formaal, Onkologisk Forskningsfond and Rigshospitalets Jubilæumsfond, Fabrikant Einar Willumsens Mindelegat; Kræftens Bekæmpelse. None of the funding sources played a role in the planning, conduction of the study, analyses of the data or writing of the manuscript.