>> Since the Medicines Market Reorganization Act (AMNOG) came into force in 2011, pharmaceutical companies have been obliged to demonstrate the additional benefit of a newly approved drug, a new combination or an already approved drug whose approval has been extended compared to one or more standard therapies. The result of this benefit assessment (NB) in accordance with Section 35a SGB V forms the basis for the subsequent negotiations on the reimbursement amount with the umbrella association of statutory health insurance companies. The central element of the NB is the benefit dossier to be submitted by the pharmaceutical company (company), in which the additional benefit compared to the so-called ACT - which may differ from the comparator in the pivotal approval studies - must be proven on the basis of patient-relevant endpoints. The NB is carried out by the Institute for Quality and Efficiency in Health Care (IQWiG) or the Federal Joint Committee (G-BA) based on the company's benefit dossier. The G-BA then decides on the recognition of an additional benefit. The methodological requirements for the NB are described in the rules of procedure of the G-BA (Federal Joint Committee 2018) and in the general methods of IQWiG (Institute for Quality and Efficiency in Health Care 2020).

In addition to mortality and morbidity, health-related quality of life (HRQoL) is a key patient-relevant target variable for assessing the therapeutic benefit of a drug (Joint Federal Committee 2018; Federal Ministry of Health 2010). The importance is also highlighted by IQWiG: “Health-related quality of life is given the same importance as serious (or severe) symptoms, subsequent complications and side effects.” (Institute for Quality and Efficiency in Healthcare 2020: 204).
Since the AMNOG came into force, HRQoL has become increasingly important in the assessment of the additional benefits of new drugs. The chairman of the G-BA, Prof. Josef Hecken, emphasizes how important quality of life (QoL) is when weighing up additional benefits (Deutscher Bundestag 2014; Staeck 2016). He values “the quality of life of patients in the last phase of life as higher than a marginal extension of life” (Staeck 2016). A lack of adequate data on HRQoL could, for example because it was not collected or because the measurement instruments used or their clinical relevance thresholds were not validated, could lead to consequences in the NB in the form of an insufficiently proven additional benefit or a time limit on the G-BA's decision (Staeck 2016; Federal Joint Committee 2018).
In the following, methodological requirements for the analysis and evaluation of QoL data in the context of NB as well as current challenges are described and discussed.

From IQWiG's perspective, when analyzing QoL data, the same methodological requirements for study design and analysis and evaluation of the data must be met as for other patient-relevant endpoints (Institute for Quality and Efficiency in Health Care 2020); The fact that the G-BA agrees with this point of view can be seen from the over 560 assessment procedures that have already been completed (as of April 2021). A methodologically sound assessment of QoL requires psychometrically validated measurement instruments (e.g. questionnaires), which include both disease-specific and generic instruments. The validation of such instruments must be documented using data that demonstrate the structural integrity of the measurement instrument in the target population and the predictive value of the measurement instrument for the patient-relevant outcome. Generic measurement instruments alone are often not suitable for recording therapy-specific effects because they are not sufficiently sensitive to reflect differences in clinical populations with often low QoL (Institute for Quality and Efficiency in Health Care 2020). In addition, they must also have been validated in the relevant disease area.
In addition to validated measuring instruments, there are also accepted relevance thresholds, e.g. Am
Form of a Minimal Important Difference (MID), required in order to be able to derive additional benefit based on QoL data. Based on these relevance thresholds, the validity of which must be proven for the current or used version of the measuring instrument, individual criteria for the existence of a therapeutic response are defined (Institute for Quality and Efficiency in Healthcare 2020). An assessment of relevance can be carried out on the basis of responder analyzes as well as standardized mean differences (SMDs, in the form of Hedges' g). Criticism of responder analyzes based on an individual MID was made in the current version 6.0 of the General Methods of IQWiG with regard to methodology and other contextual factors: The MID of a validated measuring instrument for a specific entity has not (yet) been determined empirically, which is more often the case with generic instruments (Klakow-Franck 2014), this cannot be used as a relevance threshold. It cannot be assumed that an MID valid for one entity is transferable to another patient group.
According to the IQWiG methodology, responder analyzes must be carried out for QoL data using a valid response criterion - e.g. the MID (Institute for Quality and Efficiency in Health Care 2020). If observation times vary for each patient, as is very often found in oncology studies, for example, event time analyzes that take into account the time until an event occurs (Unkel et al. 2019) are viewed as the appropriate methodology.
Depending on the entity and stage of the disease, an event is considered for a patient to be, for example, the improvement or deterioration of a scale on the measuring instrument, defined as a change that is sufficiently noticeable for the patient compared to the initial value at study inclusion. There is currently no established standard that can be used to estimate the significance of the MIDs used. For a first proposal published in 2020 for an instrument for the quality assessment of studies to determine MIDs and their significance, comprehensive results are still missing (Devji et al. 2020) and the criteria used there to assess the quality of anchors are used to determine threshold values not viewed uncritically by experts (Peipert 2020). The IQWiG currently defines - instead of previously a change of at least the validated MID - a value of 15% of the scale range as a plausible threshold for a small but sufficiently noticeable change (Institute for Quality and Efficiency in Health Care 2020).
If the prespecified response criterion in the sense of an MID is below 15% of the scale range, in addition to a post-hoc definition of the response criterion of exactly 15% of the scale range, continuous models (e.g. mixed models for repeated measurements) with the addition of normative effect estimators such as Hedges' g can be used. Knowing the temporal stability of the scale in the target population, such normative effect estimates can in principle be used to estimate the distribution of change measures (Morris & DeShon 2002), which can also be used to discuss response rates based on possible MID scenarios.
The return of the questionnaires should be sufficiently large for the data to be usable. As a rule, a threshold of 70% is set here, i.e. the results should be based on at least 70% of the eligible study participants. In a second step, IQWiG currently differentiates the follow-up losses and describes that these can, if necessary, be compensated for through adequate replacement strategies (Institute for Quality and Efficiency in Health Care 2020).
The statistical evaluation of QoL data in the context of studies with a survival time endpoint places special demands on the statistical analysis, among other things due to different lengths of follow-up periods and censoring.
The aspects of the analysis and evaluation of QoL data described in more detail below represent a selection of important and currently most discussed sub-areas in the context of NB. Other relevant aspects, such as the validity of the measurement instruments, floor and ceiling effects, magnitude of the effect (beyond... Relevance threshold) are not examined in more detail here, but should be taken into account as part of a further methodological discussion.

When assessing the relevance of observed changes on a QoL scale, not only their statistical significance but also, in particular, their clinical relevance must be taken into account. This requires the definition of a threshold value, commonly referred to as the response threshold, as a criterion for a relevant change on the QoL scale.
While a response threshold does not necessarily represent a minimum threshold for relevant changes, the MID is the value that defines the smallest clinically relevant difference between time points. However, defining the MID presents challenges.
A first problem is the inconsistent use of the term MID. In randomized clinical trials in which groups of patients (= the study arms) are compared with one another, the smallest clinically relevant difference between groups would have to be determined. However, the “relevance” of the difference is difficult to assess. If you refer to external criteria such as functionality or toxicity profiles, you compare different constructs. The admissibility of the content is questionable because external criteria are external assessments by the doctor, but QoL is by definition a subjective construct. Using the doctor's external assessment as an external criterion for QoL would only be permissible if it could be proven that it corresponds to the patient's self-assessment, which is usually not the case. Therefore, a second approach is usually chosen: the Q‑Q scores of patients at two points in time are compared with each other. In addition, patients are asked to what extent their quality of life has improved or worsened. The mean difference in QoL scores among patients who reported minimal changes in QoL is the MID. Here you check the smallest clinically relevant difference between time points. Since the comparison between study arms and the comparison between points in time are conceptually fundamentally different, a distinction should also be made in terms of terminology. Crosby et al. (2003) proposed the term MID for the former and the term “Clinically Meaningful Change” (CMC) for the latter (Crosby et al. 2003).
Further problems are, on the one hand, that it has been empirically shown that the CMC differs when QOL worsens versus improves (Cocks et al. 2012). On the other hand, MID and CMC that were determined for one patient group do not automatically have to apply to another group.
In order to reflect the variability of an MID of a survey instrument with regard to the type and severity of an illness, the direction of change being examined or the methodology, IQWiG has developed a new approach in its current methodology that is intended to pragmatically enable the determination of thresholds that are sufficiently certain delineate a clinically significant area. Specifically, the institute specifies that in the future it will use responder analyzes from a limit value of at least 15% of the scale range of the survey instrument used for the evaluation (Institute for Quality and Efficiency in Health Care 2019). The chosen threshold value is justified by the fact that a review of systematic reviews on MIDs has shown that in many cases MIDs range between 10 and 20% of the range of scores of a survey instrument.
According to IQWiG, by setting a threshold of 15%, the aim is to ensure that a sufficiently safe change that is noticeable for patients can be depicted and that the risk of results-driven reporting is minimized by selecting one of many possible MIDs. This new approach enables predictability of the acceptance of response thresholds and also offers a way to detect relevant changes in QoL for survey instruments for which there is no validated MID in the corresponding indication.
However, this new pragmatic approach from IQWiG is also subject to some limitations. There are survey instruments that do not have a clearly defined scale and the scores of their results must first be converted using a T-score in order to be able to classify the changes (e.g. SF-36). In addition, there is no consensus on how a threshold of 15% should be determined. The maximum and minimum computable scores or external (i.e. empirical) values can be used to define the range. Alternatively, the range can be defined based on a fixed number of standard deviations from the mean.
Another limitation is that no patient assessment is included in the validation of the 15% threshold. One way to incorporate the patient's view of the effectiveness of an intervention with regard to QoL would be for patients to self-assess the threshold at which one can speak of a “sufficiently good” or a “relevantly bad” QoL. The concept of such threshold values (cut-offs) is well established in other areas in which patient-reported outcomes (PRO) are presented (Zigmond & Snaith 1983). In the case of QoL, it seems important that only people who have experienced the disease and thus the possible QoL impairment themselves can meaningfully assess this threshold. This approach has been shown to be feasible in pilot studies (Fahsl et al. 2012).
The first NB procedures according to the new IQWiG methodology show that while the IQWiG already strictly stipulates the 15% limit for the recognition of responder analyses, the G-BA is currently also recognizing evaluations according to the old methodology. In the NB procedure for secucinumab in the indication psoriatic arthritis, for example, the HAQ-DI questionnaire was used to record disability caused by inflammatory rheumatic joint disease. IQWiG did not take into account the MID of ≥3.5 initially presented by the company with reference to the new methods for the NB and only used the analysis subsequently submitted by the company as part of the statement with a response threshold of 15% (corresponding to an MID of ≥0.45 ).
In the present indication, the G-BA had already recognized the MID of ≥0.35 points as a clinically relevant change in the HAQ-DI and, against the background of the current methodological discussion, used both responder analyzes to assess the additional benefit. The same applies to the SF-36 questionnaire in the same NB procedure for measuring HRQoL in clinical studies (Joint Federal Committee 2020).
As part of consultations and in the FAQs published on its website (Joint Federal Committee 2020, 2021a, 2021b), the G-BA currently refers to the “General Methods” 6.0 of the IQWiG for the implementation and evaluation of the clinical relevance thresholds. During a transitional period that applies until an adapted template for Module 4 of the benefit dossier comes into force, the companies are required to carry out responder analyzes using a clinical relevance threshold that has already been accepted by the G-BA in the past (e.g. a MID) as well as the 15% threshold (Joint Federal Committee 2021a).
On December 16, 2021, at its public meeting, the G-BA decided to change the module templates in Appendix II to Chapter 5 of the Rules of Procedure (Joint Federal Committee 2021b). Changes are hereby made that take up the new IQWiG specifications for MID-based responder analyzes at PRO.

A sufficient return of questionnaires ensures that the sample surveyed is representative of the study population. If rejection rates are too high, an increased selection bias can be expected.
The terms must be clearly separated here: rejection rate defines the proportion of patients who already refuse to participate in the survey at baseline. Dropout rate defines the proportion of patients who initially participate in the study but no longer participate at follow-up points. This can result in a selection bias if patients with poorer QoL drop out of the study more often, but this is more the rule than the exception. It is therefore important to achieve the lowest possible rejection and dropout rates. But what exactly does “as little as possible” mean?
In its method paper 6.0, IQWiG describes the assessment of missing values in clinical studies - here missing questionnaires - in a two-stage process. First, study participants who were completely excluded from the analysis (= rejection rate) are considered. Results should be based on at least 70% of the patients included in the analysis in order to be considered for the NB. In the following analysis of follow-up losses (= dropout rate), IQWiG does not set any fixed limits, but depends on the following factors depending on the context: 1. number and timing of follow-up losses and 2. reasons for the follow-up -Losses (Institute for Quality and Efficiency in Healthcare 2020).
However, experience from a large number of NB procedures shows that the reference value is an essential question and difficulty in determining and evaluating dropout rates. To determine the dropout rates, IQWiG and G-BA take all randomized patients into account in the denominator and only deduct deceased patients, which means that patients who, according to the study protocol, no longer receive a questionnaire because they have finished the treatment (in advance) and the QoL only recorded until the end of treatment are counted as “dropouts”. This contrasts with the company's opinion that patients in this constellation are no longer in the risk set after the end of treatment and therefore cannot be assessed as “dropouts”. Both refusal and discontinuation rates should be based on the patients available in the analysis set at the respective time. It could also be informative and relevant to compare this parameter with those previously used by IQWiG and G-BA.
Comment on the statistical

In oncology in particular, QoL data is collected as part of survival studies. Studies with a survival time endpoint are not performed due to different lengths of follow-up times and censoring using statistical methods of survival time analysis (especially chapter
lan-Meier, Cox's proportional hazards model); this also makes the analysis of QoL more difficult. Survival time analysis techniques are often used to analyze the time until deterioration, although the reasons for censoring can include progression or death, which represent competing risks. Competing risk censoring is subtle: the Cox model provides a correct hazard ratio (HR) for time to deterioration. However, censoring is informative with regard to probability statements and must therefore be supplemented by the ability to simultaneously analyze multiple event times of competing risks. Kaplan-Meier curves are also not useful because they overestimate the actual frequencies. Instead, the Aalen-Johansen estimator of the so-called cumulative incidence function should be used (Allignol et al. 2016; Beyersmann & Schmoor 2019).
“Censorship through progression” plays a special role. In this case, the study protocol often stipulates that QoL data will no longer be collected. This is often associated with the criticism that progression is also an informative censorship regarding HR. However, this is not the case for the analysis of HR in the context of competing risks and for the duration of progression-free survival.
One point of criticism of the “time to deterioration” target is that it does not capture a possible improvement in QoL. However, an analysis of the duration of a relatively good QOL is possible by extending competing risks to multistage models, which was already proposed in the early 1990s (Olschewski et al. 1994; Olschewski & Schumacher 1990; Schumacher et al. 1991). Current developments link multi-stage models, censoring when changing treatment (e.g. after progression) with causal analyzes (Gran et al. 2015) and are part of the current Estimand debate (Unkel et al. 2019). An expansion of the discourse to include QOL measurement would also be desirable.
The analyzes discussed so far require that QoL be categorized, which facilitates both statistical analysis and its interpretation. However, to directly examine the longitudinal course of QoL, mixed linear models are often used. An implicit assumption here is that QoL data that are not available are “missing at random,” meaning that whether data is missing or not is independent of the extent of the missing values. However, if the longitudinal QoL trajectory is to be evaluated as such, an alternative to the mixed linear model is a joint model in which the “informative drop out” process is also modeled and can therefore be “missing not at random” (Rizopoulos 2012). . The price for this is typically relatively strict model assumptions and, often, numerical difficulties. Furthermore, a joint model even allows the questionable extrapolation of QoL after death.

The assessment of patients' HRQoL in clinical studies is based on subjective assessments by the patient. An important goal when evaluating the data obtained is to still obtain the most reliable insights possible. This concern will become even more urgent with the introduction of the NB of medicines as part of the AMNOG, as the focus here is exclusively on patient-relevant endpoints. The current examples discussed here illustrate the growing importance of QoL data in the NB process, but also highlight which methodological and conceptual problems and challenges arise from this and which solutions are currently being discussed. There is an acute need for new approaches that should, if possible, take into account all the interests involved. A constant exchange between academic research, the institutions responsible for the assessments (G‑BA; IQWiG) and the companies with the aim of finding such solutions is more than desirable - especially in the interests of patients.
There are positive approaches in this direction: the work of the project group “Analysis of adverse events with variable observation times”, the working group “Therapeutic Research” (ATF) of the German Society for Medical Informatics, Biometry and Epidemiology eV (GMDS) and the working group are mentioned here “Pharmaceutical Research” (APF) of the German Region of the International Biometric Society (IBS-DR).
It is to be welcomed that an exchange regarding the analysis and evaluation of PRO is becoming increasingly important and that the development of solution approaches is being promoted. In 2020, for example, the challenges in analyzing PRO were the topic of a workshop by the ATF of the GMDS and in June 2021 the focus of the “IQWiG in Dialogue” event series. The interdisciplinary setting in-
ternational Standards in Analyzing Patient-Reported Outcomes and Quality of Life Endpoints Data (SISAQOL) Consortium published its international recommendations for the analysis of PROs in February 2020 (Coens et al. 2020), which also help in the development of standards in the analysis of PROs should be taken into account in the NB. <<

from:
Sarah Böhme MSc (1)
Prof. Dr. rer. med. Susanne Singer (2)
Stephanie Kauffmann MSc (3)
Dipl.-Stat. Friedhelm Leverkus (4)

1: Manager Health Technology Assessment at Pfizer in Germany
2: Head of the Epidemiology and Health Care Research Department at the Institute for Medical Biometry, Epidemiology and Informatics at the Mainz University Medical Center
3: Junior Manager Health Technology Assessment at Pfizer in Germany
4: Director Health Technology Assessment & Outcomes Research Health & Value at Pfizer in Germany

1. Federal Joint Committee (G-BA) (2018): Rules of Procedure of the Federal Joint Committee. Last changed on August 16, 2018 - published in the Federal Gazette BAnz AT March 5, 2019 B2 - came into force on March 6, 2019. In: https://www.g-ba.de/downloads/62-492-1777/VerfO_2018 -08-16_iK-2019-03-06.pdf. (accessed on: June 17, 2021)
2. Institute for Quality and Efficiency in Healthcare (IQWiG) (2020): General methods. Version 6.0 from November 5th, 2020. In: https://www.iqwig.de/methods/generale-methods_version-6-0.pdf?rev=144030. (accessed on: January 18, 2021)
3. Federal Ministry of Health (2010): Drug Benefit Assessment Ordinance - AM-NutzenV. In: http://www.gesetze-im-internet.de/am-nutzenv/AM-NutzenV.pdf. (accessed on: June 17, 2021)
4th German Bundestag (2014): Hecken praises the pharmaceutical law. In: Today in the Bundestag 2014, 145. In: https://www.bundestag.de/webarchiv/Presse/hib/2014_03/01-262758. (accessed on: August 25, 2021)
5. Staeck, F. (2016): GBA boss wants priority for quality of life. In: Ärzte Zeitung, June 3, 2016, http://www.aerztezeitung.de/politik_gesellschaft/arzneimittelpolitik/nutzenvergleich/article/9 12741/nutzenvergleich-gba-chef-will-vorfahrt-lebensqualitaet.html. (accessed on: August 12, 2021)
6. Federal Joint Committee (G-BA) (2018): Main reasons for the Federal Joint Committee's decision to amend the Medicines Directive (AM-RL): Annex XII - Extension of the period of validity of a decision on the benefit assessment of medicines with new ones Active ingredients according to § 35a SGB V – Trifluridine/Tipiracil. In: https://www.g-ba.de/downloads/40-268-5105/2018-07-05_AM-RL-XII_Trifluridin-Tipiracil_Verlaengerung-Beszeitung_D-252_TrG.pdf. (accessed on: June 17, 2021)
7. Klakow-Franck, R. (2014): The importance of quality of life for the work of the Federal Joint Committee. In: ZEFQ 2014, 108, 2: 151-6
8. Unkel, S., Amiri, M., Benda, N., Beyersmann, J., Knoerzer, D., Kupas, K., et al. (2019): On estimates and the analysis of adverse events in the presence of varying follow-up times within the benefit assessment of therapies. In: Pharm Stat 2019, 18, 2: 166-83
9. Devji T, Carrasco-Labra A, Qasim A, Phillips M, Johnston BC, Devasenapathy N et al. (2020): Evaluating the credibility of anchor based estimates of minimal important differences for patient reported outcomes: instrument development and reliability study. In: BMJ 2020, 369: m1714
10. Peipert, JD (2020): Re: Evaluating the credibility of anchor based estimates of minimal important differences for patient reported outcomes: instrument development and reliability study. In: BMJ 2020, 369:m1714
11. Morris, SB, DeShon, RP (2002): Combining effect size estimates in meta-analysis with repeated measures and independent-groups designs. In: Psychological Methods 2002, 7, 1: 105-25
12. Crosby RD, Kolotkin RL, Williams GR (2003) Defining clinically meaningful change in health-related quality of life. In: J Clin Epidemiol 2003, 56, 5: 395-407
13. Cocks K, King MT, Velikova G, de Castro G Jr, Martyn St-James M, Fayers PM et al. (2012): Evidence-based guidelines for interpreting change scores for the European Organization for the Research and Treatment of Cancer Quality of Life Questionnaire Core 30. In: Eur J Cancer 2012, 48, 11: 1713-21
14. Institute for Quality and Efficiency in Healthcare (IQWiG) (2019): General methods. Draft for version 6.0 from December 5th, 2019. In: https://www.iqwig.de/download/Allgemeine-Methods_Entwurf-fuer-Version-6-0.pdf. (accessed on: June 17, 2021)
15. Zigmond AS, Snaith RP (1983) The hospital anxiety and depression scale. In: Acta Psychiatr Scand 1983, 67, 6: 361-70
16. Fahsl, S., Keszte, J., Boehm, A., Vogel, HJ, Völkel, W., Meister, EF et al. (2012): Clinical relevance of quality-of-life data in laryncectomized patients. In Laryngoscope 2012, 122, 7:1532-8
17th Federal Joint Committee (G-BA) (2020): Benefit assessment procedure for the active ingredient secukinumab (new scientific findings (§ 14): psoriatic arthritis). In: https://www.g-ba.de/vergleichverfahren/nutzenvaluation/590/#nutzenvaluation. (accessed on: June 17, 2021)
18th Federal Joint Committee (G-BA) (2021): Answers to frequently asked questions about the benefit assessment process. Available at: https://www.g-ba.de/themen/arzneimittel/arzneimittel-politik-anlagen/nutzenvaluation-35a/faqs/#wie-soll-vor-dem-background-der-veroffentlichung-des-methodologies-60 -of-the-iqwig-on-november-5-2020-currently-in-the-dossier-preparation-the-determination-of-clinical-relevance-thresholds-for-complex-scales-are-dealt-with (accessed August 12, 2021)
19th Federal Joint Committee (G-BA) (2021): Decision of the Federal Joint Committee on a change to the rules of procedure: Change to the module template in Appendix II to Chapter 5 of December 16, 2021. In: https://www.g- ba.de/downloads/39-261-5217/2021-12-16_VerfO_Aenderung-Modulvorlage-Annex-II-Kap-5.pdf. (accessed on: December 27, 2021)
20. Allignol, A., Beyersmann, J., Schmoor, C. (2016): Statistical issues in the analysis of adverse events in time-to-event data. In: Pharm Stat 2016, 15, 4: 297-305
21. Beyersmann, J., Schmoor, C. (2019): The Analysis of Adverse Events in Randomized Clinical Trials. In: Halabi, S., Michiels, S. (eds.) (2019): Textbook of Clinical Trials in Oncology - A Statistical Perspective. 1st edition. New York: Chapman and Hall/CRC, 537-557
22. Olschewski, M., Schulgen, G., Schumacher, M., Altman, DG (1994): Quality of life assessment in clinical cancer research. In: Br J Cancer 1994, 70, 1: 1-5
23. Olschewski, M., Schumacher, M. (1990): Statistical analysis of quality of life data in cancer clinical trials. In: Statistics in Medicine 1990, 9, 7: 749-63
24. Schumacher, M., Olschewski, M., Schulgen, G. (1991): Assessment of quality of life in clinical trials. In: Statistics in Medicine 1991, 10, 12: 1915-30
25. Gran, JM, Lie, SA, Øyeflaten, I., Borgan, Ø., Aalen, OO (2015): Causal inference in multi-state models–sickness absence and work for 1145 participants after work rehabilitation. In: BMC Public Health 2015, 15, 1: 1082
26. Rizopoulos, D. (2012): Joint Models for Longitudinal and Time-to-Event Data: With Applications in R. New York: Chapman and Hall/CRC
27. Coens, C., Pe, M., Dueck, AC, Sloan, J., Basch, E., Calvert, M., et al. (2020): International standards for the analysis of quality-of-life and patient-reported outcome endpoints in cancer randomized controlled trials: recommendations of the SISAQOL Consortium. In: Lancet Oncol 2020, 21, 2: e83-e96