Study endpoints in the early benefit assessment from the perspective of IQWiG

Patient relevance is firmly anchored in the early benefit assessment. According to the German Pharmaceutical Benefit Ordinance (AM-NutzenV) ¹ , the benefit of a medicinal product is defined as "the patient-relevant, therapeutic effect, particularly with regard to improving health, shortening the duration of illness, prolonging survival, reducing side effects, or improving quality of life. The added benefit within the meaning of this ordinance is a benefit that is quantitatively or qualitatively greater than the benefit provided by the appropriate comparator therapy" (Section 2, Paragraph 4 AM-NutzenV) ¹ . According to the IQWiG methodology paper ² , patient relevance is "how a patient feels, can perform their functions and activities, or whether they survive."

The focus of the benefit assessment is therefore on endpoints that can be perceived directly by patients. In the early benefit assessment, patient-relevant endpoints are assigned to the four categories of mortality, morbidity, health-related quality of life, and adverse effects. These must be distinguished from endpoints that are based, for example, on imaging or laboratory findings and are not noticeable to patients. Changes in the dosage form are also repeatedly proposed as a patient-relevant benefit per se. Without evidence that a new dosage form influences patient-relevant endpoints such as health-related quality of life, such a benefit cannot be derived either according to the Social Code Book V or from the IQWiG definition of patient relevance. A patient-relevant benefit can only be derived in exceptional cases, e.g. when oral administration is possible instead of intrathecal administration, ³ and thus a clear reduction in complications of medication administration can be assumed.

High priority given to patient-reported endpoints

Patient-reported outcomes (PROs) are of great importance for morbidity and quality of life endpoints; their role in clinical trials has been strengthened by their significance in the AMNOG process. PROs allow patients to assess how the use of a new medication affects their symptoms or health-related quality of life, for example. Without PROs, only an incomplete picture of the benefits and harms of new medications emerges. Patient involvement is essential even during the development of PRO instruments. This is the only way to ensure that all aspects relevant from the patient's perspective are presented clearly and comprehensively.

PROs are generally preferable to anthropometric parameters such as height as a patient-relevant endpoint. Although height (z-score) was classified as patient-relevant in the indication of achondroplasia ⁴ , it is difficult to estimate how a specific change in the height endpoint (z-score) will ultimately affect the patient, for example, on their functional limitations and pain. Therefore, the added benefit in the height endpoint (z-score) could not be conclusively quantified for the benefit assessment.

The use of PROs is also possible for endpoints on adverse events, although they are not yet standard practice. Furthermore, not all adverse events are suitable for recording using PROs; for example, directly observable/measurable events (e.g., retinal tears) are unsuitable. Subjectively perceived adverse events such as nausea, however, can certainly be recorded using PROs. For this purpose, the National Cancer Institute (NCI) has developed the PRO-Common Terminology Criteria for Adverse Events (PRO-CTCAE) system for assessing symptomatic toxicity in patients in oncology studies. ⁵ The PRO-CTCAE system has already been submitted to IQWiG in several benefit assessments, for the first time in A20-87 and A23-86. However, the corresponding data were not usable because, first, there was no detailed justification for the selection of the symptomatic AEs used from the PRO-CTCAE system, and, second, the results were presented only descriptively without taking into account different observation periods. ^6,7

Only a few surrogate validations have been presented

In early benefit assessments, IQWiG has so far accepted three surrogate endpoints as sufficiently valid surrogates for different patient-relevant endpoints: virological response as a surrogate for AIDS/overall survival in HIV infection, _HbA1c value as a surrogate for microvascular complications in type 1 diabetes, and sustained virological response (SVR) as a surrogate for preventing the occurrence of hepatocellular carcinoma in hepatitis C infection. ^8-10 Overall, however, only a few validations of surrogates for the early benefit assessment were submitted to IQWiG, and these were primarily in the oncology field (see Table 1).

Of particular note here is progression-free survival (PFS), for which surrogate validations have been presented most frequently (for the indications melanoma, breast cancer, and prostate cancer ^11-15 ). Surrogate validations have also been presented for disease-free survival (DFS) in adjuvant breast cancer, metastasis-free survival in prostate cancer, and annual forced vital capacity in interstitial lung disease. ^15-18

However, the validation studies presented were not suitable for demonstrating the suitability of the respective surrogate endpoints for overall survival. Regardless of the question of suitability as a surrogate for overall survival, some endpoints or operationalizations that reflect cancer progression were classified as patient-relevant: for example, symptomatic progression in prostate cancer, which was operationalized via symptoms noticeable to the patient, was accepted in two benefit assessments. ^{15, 19} Recurrences were also generally accepted as a patient-relevant endpoint in various benefit assessments. ^{16, 17}

Although the assessment of the endpoint "recurrence" is based on imaging, it is considered a failure of the curative approach if the tumor is detectable again during or after (adjuvant) therapy with curative intent. For patients, this may represent a transition to an incurable stage of the disease and is therefore directly relevant to the patient. Furthermore, the failure of the curative treatment approach is also relevant for patients who are not tumor-free at baseline and has been used by the IQWiG as an independent endpoint in hemato-oncology patients with diffuse large B-cell lymphoma (DLBCL) ^20-22 , but also in patients with solid tumors ^23,24 .

Methodological requirements for valid surrogates

Surrogate validations are generally possible based on randomized controlled trials (RCTs) and in other specific situations (such as those encountered in SVR). The former generally require a meta-analysis of multiple RCTs that investigated both effects on the surrogate endpoint and effects on the patient-relevant endpoint of interest. Recognized validation methods are correlation-based methods, such as examining the correlation between effects at the study level and the surrogate threshold effect (STE).

In addition to the methodology used for surrogate validation, it is of fundamental importance that the underlying study pool is complete and suitable. The validation studies presented in the benefit assessments for pertuzumab (breast cancer) ^{16, 17} and nintedanib (interstitial lung disease) ¹⁸ show that it is fundamentally possible to conduct surrogate validation in accordance with IQWiG requirements. In the project on pertuzumab, the DFS and, for nintedanib, the annual forced vital capacity were each to be shown as suitable surrogates for overall survival. For pertuzumab, the underlying study pool of the validation study was unsuitable because it excluded studies that would have been relevant in the therapeutic indication presented in this benefit assessment. For nintedanib, the methodological implementation of the validation was flawed, which led to an underestimation of the STE. After self-calculated correction, it was shown that the effect on the surrogate was not large enough to derive an effect on overall survival.

In addition, there are special situations in which validity can also be recognized. ²⁵ This requires that the relationship between the patient-relevant endpoint and the surrogate endpoint is clearly biologically/medically plausible and that other criteria are met. One example is the SVR in patients with chronic hepatitis C infection, where the occurrence of the surrogate endpoint led to a significantly reduced risk of hepatocellular carcinoma. In addition, the risk with regard to the actual endpoint reached a minimal level, namely that of a non-affected population. ¹⁰ In these special situations, the data basis must be cohort studies that refer to people under treatment and whose follow-up period is sufficiently long to adequately capture the risk of the actual endpoint occurring.

Benefit assessment and treatment decisions require sufficiently reliable data

From IQWiG's perspective, it is problematic that regulatory authorities often accept surrogate endpoints for the approval of new drugs that are not sufficiently validated, and that the use of such surrogate endpoints has increased over the past decades, both at the Food and Drug Administration (FDA) and the European Medicines Agency (EMA). ^26-28 For (possibly accelerated/conditional) approval, the position is often taken that this is justifiable and that one is willing to accept the greater uncertainty of results in order to make a new drug available quickly.

However, accepting a high degree of uncertainty at the time of approval must not lead to a situation where truly reliable data are never generated. The question of benefit assessment and evidence-based treatment decisions require sufficient certainty. The goal of early benefit assessment is to filter out those new drugs with a "proven additional benefit" rating that are sufficiently certain to provide added value for patients, rather than merely "perhaps."

Evidence-based medical care also requires meaningful data for well-founded treatment decisions. This requires meaningful endpoints and, in certain cases, more or better surrogate validations. Furthermore, the frequently voiced accusation that treatment parameters described in guidelines and disease management programs (DMPs) are ignored in benefit assessments also misses the core of the problem: The use of a laboratory parameter recommended in a guideline or DMP to guide therapy in an individual patient by no means necessarily legitimizes its suitability as a patient-relevant endpoint in a clinical trial. The decisive factor for the benefit assessment can only be whether there is sufficient certainty that an effect in the surrogate endpoint will also be reflected in an effect in the patient-relevant endpoint.

Data on chronic diseases remains poor

Particularly in the area of ​​chronic diseases, there is often a significant lack of suitable data for early benefit assessments. From 2021 to the beginning of August 2024, IQWiG conducted a total of 169 benefit assessments in the area of ​​chronic diseases. Suitable data were not available for 68 percent (115) of the projects (see Figure 1).

The problem here is not the endpoints, but rather the study design. Even though the majority of studies were RCTs, the data are often unsuitable for benefit assessment because these RCTs are based on a comparison with a placebo or an inadequate therapy in the comparator arm, rather than a direct comparison with the current standard therapy. In other cases, no comparison is found at all (single-arm studies). Data from such studies are generally unsuitable for benefit assessment. As early as 2019, it was criticized that suitable data are often not available for benefit assessment. ²⁹ Unfortunately, with regard to chronic diseases, no positive development can be seen in this regard. The reasons for this stagnation and possible starting points for change (targeted use of positive or negative stimuli) should be discussed.

Of the 54 (32 percent) benefit assessments conducted between 2021 and the beginning of August 2024 in which suitable data for the benefit assessment were available, over 70 percent of the assessments relied on approval studies. In the majority of these cases, only one approval study was available for the benefit assessment. Non-approval studies (Figure 1) were used for 15 (28 percent) of the benefit assessments. Overall, the primary study endpoint was fully patient-relevant in only about a quarter of the included studies. In over 40 percent of the included studies, the primary endpoint was not patient-relevant because, as described above for approval studies, it was based, for example, on an (insufficiently validated) surrogate endpoint (e.g., change in _HbA1c in patients with type 2 diabetes or the eGFR slope in patients with Fabry disease, see Figure 2).

Regarding eGFR, it should be noted that changes in eGFR are only patient-relevant if it is sufficiently certain that the eGFR falls to a level perceptible to the patient. This was the case for the first time in one of the two addenda to Finerenone ^30. The combined endpoint of the submitted studies on renal morbidity included the individual components of renal failure (defined as a confirmed sustained decrease in eGFR to < 15 ml/min/1.73 m2 or end-stage renal disease), eGFR decrease ≥ 57 percent, and renal-related death. Based on the mean baseline eGFR values ​​(approximately 43 ml/min/1.73 m2) of the patients, sufficient patient relevance of the component eGFR decrease ≥ 57 percent could be assumed in the present data situation, and the combined endpoint was used.

In 30 percent of the included studies, the primary endpoint could only be used with limitations (e.g., change in average monthly migraine days from baseline vs. 50 percent reduction in migraine days/month, or only individual components of a combined endpoint were used, see Figure 2). The operationalization of an endpoint can therefore also be crucial for whether an endpoint is considered patient-relevant or not.

It is therefore clear that the fact that an endpoint was investigated in (approval) studies does not necessarily mean that it is a patient-relevant endpoint. Instead of "study endpoint -> relevant," should "relevant endpoint -> study endpoint" apply, as suggested by Thomas Kaiser in 2016? ³¹ In principle, HTA decisions should not incentivize the use of study endpoints that entail high uncertainty and instead reward the investigation of meaningful, patient-relevant endpoints. ²⁹

However, the question of the patient relevance of the primary endpoint or its operationalization is rarely directly relevant to the conclusion regarding the question "Is there an additional benefit: yes or no?" Only in six projects (11 percent) from 2021 to the beginning of August 2024 was no additional benefit derived, although the submitted studies showed an advantage for the intervention in the primary (but not patient-relevant) endpoint (e.g., change in _HbA1c value in type 2 diabetes). However, there are usually other relevant endpoints that showed an additional benefit (this was the case in 24 projects [44 percent]). One example is the evaluation of risankizumab, in which the primary endpoint (endoscopic remission) was not accepted, but positive effects were shown in the Inflammatory Bowel Disease Questionnaire (IBDQ; total score and subscore bowel symptoms) and Short Form-36 (SF-36). ³²

In other assessments, negative effects were also observed alongside positive effects, which, overall, led to the conclusion "no added benefit" (two projects [4 percent]). In seven projects (13 percent), the primary (non-patient-relevant) endpoint showed no effect, and the rejection therefore had no influence on the conclusion of the benefit assessment (see also Figure 3). These studies were non-inferiority studies. However, it is problematic if, in these cases, there are de facto no meaningful endpoints on mortality, morbidity, or quality of life, i.e., benefit endpoints. This was the case, for example, with vadadustat ³³ , but the most recent diabetes assessments (type 2) also show that the studies are still not focused on the treatment goals. ³⁴

Improvements in PRO analyses necessary

The data quality of PROs for benefit assessments is often insufficient, so that only a portion of the submitted PRO analyses are ultimately suitable for benefit assessment. ³⁵ Another common problem is that the PROs are collected for too short a time. This is probably because the studies submitted for benefit assessments are heavily tailored to approval. For example, the PROs are only recorded up to disease progression. But for those affected, symptoms and quality of life are of course also relevant after the disease has worsened. Whether the patient fares better or worse in the long term with the intervention being evaluated than with the comparator treatment can only be assessed if the corresponding PROs are collected beyond the progression. ³⁶

It is also particularly important that the PROs are collected in full during the survey period. Low response rates to the relevant questionnaires often mean that the data cannot be used for benefit assessment. In a discussion round of various stakeholders, patient representatives also expressed that those affected are quite willing to invest time in PRO collection if they are specifically explained why their information is needed. This shows that a longer survey of such outcomes is fundamentally feasible – and in no way contradicts the interests of those affected. ³⁷

The analysis and operationalization of PRO endpoints must be carefully discussed on a case-by-case basis, as numerous aspects can influence the significance of the results. For example, depending on the indication and therapeutic goal, (sustainable) improvement may be of primary relevance in some cases, while deterioration may be of primary relevance in others. Should continuous evaluations or responder analyses at a specific evaluation point in time be considered? Which point in time would be most suitable for a responder analysis? ³⁸ Which response criterion is suitable for indicating a noticeable change? For the question of the response criterion, the patient-specific change of 15 percent of the range of the survey instrument proposed by IQWiG has proven to be practical. ³⁹ The aim of this approach developed by IQWiG was also to create clarity for manufacturers and prevent outcome-driven reporting. Since its introduction, this response threshold has been regularly used by manufacturers in studies and dossiers.

Conclusion

IQWiG principles remain unchanged: patient relevance is the paradigm of benefit assessment.

Current status of surrogates: Few validations have been submitted, primarily in the oncology field. Numerous publications in recent years show that regulatory authorities accept surrogates whose validity has not been sufficiently demonstrated – more validations (where necessary) are desirable.

Instead of “study endpoint–> relevant” the term “relevant endpoint–> study endpoint” should apply.

Improvements to PRO analyses necessary:

• fundamental methodological quality (completeness and quality of data) and duration of the PRO survey often need improvement (e.g. beyond Progress).
• The usefulness of analyses must be discussed on a case-by-case basis (e.g. symptom burden, [permanent] improvement/deterioration; handling of intercurrent events; time point considered).
• Overall, (even) more high-quality data on PROs is needed.

literature

¹ Federal Ministry of Health (2023) Ordinance on the Benefit Assessment of Medicinal Products pursuant to Section 35a Paragraph 1 SGB V for Reimbursement Agreements pursuant to Section 130b SGB V (Medicinal Products Benefit Assessment Ordinance - AM-NutzenV). https://go.sn.pub/f5cp5w. Accessed 09.10.2023.

² Institute for Quality and Efficiency in Health Care (2023) General Methods; Version 7.0. https://go.sn.pub/mkwigq. Accessed October 6, 2023.

³ Institute for Quality and Efficiency in Health Care (2021) Risdiplam (spinal muscular atrophy) – Benefit assessment according to Section 35a of the Social Code Book V; Dossier assessment. https://go.sn.pub/doprfg. Accessed July 11, 2023.

⁴ Institute for Quality and Efficiency in Health Care (2024) Vosoritide (achondroplasia); Addendum to Project A23-92 (dossier assessment). https://doi.org/10.60584/A24-08. Accessed 23.02.2024.

⁵ Institute NC Patient-Reported Outcomes version of the Common Terminology Criteria for Adverse Events (PRO-CTCAE). https://go.sn.pub/n7lwyc. Accessed October 7, 2024.

⁶ Institute for Quality and Efficiency in Health Care (2020) Durvalumab (small cell lung cancer) – Benefit assessment according to Section 35a of the Social Code Book V; Dossier assessment. https://go.sn.pub/90yg37. Accessed July 11, 2023.

⁷ Institute for Quality and Efficiency in Health Care (2023) Sacituzumab govitecan (breast cancer); Benefit assessment according to Section 35a of the Social Code Book V; Dossier assessment. https://doi.org/10.60584/A23-86. Accessed November 17, 2023.

⁸ Institute for Quality and Efficiency in Health Care (2012) Rilpivirine – Benefit Assessment according to Section 35a of the Social Code Book V; Dossier Assessment. https://go.sn.pub/zphzud. Accessed July 11, 2023.

⁹ Institute for Quality and Efficiency in Health Care (2015) Insulin degludec (new therapeutic indication) – Benefit assessment according to Section 35a SGB V; Dossier assessment. https://go.sn.pub/3lai7e. Accessed July 11, 2023.

¹⁰ Institute for Quality and Efficiency in Health Care (2011) Boceprevir – Benefit Assessment according to Section 35a of the Social Code Book V; Dossier Assessment. https://go.sn.pub/0605wp. Accessed July 11, 2023.

¹¹ Institute for Quality and Efficiency in Health Care (2013) Dabrafenib – Benefit Assessment according to Section 35a of the Social Code Book V; Dossier Assessment. https://go.sn.pub/qnv9l4. Accessed July 11, 2023.

¹² Institute for Quality and Efficiency in Health Care (2015) Pembrolizumab – Benefit Assessment according to Section 35a of the Social Code Book V; Dossier Assessment. https://go.sn.pub/0lfwif. Accessed July 11, 2023.

¹³ Institute for Quality and Efficiency in Health Care (2017) Palbociclib (breast cancer) – Benefit assessment according to Section 35a of the Social Code Book V; Dossier assessment. https://go.sn.pub/0r0kkc.. Accessed July 11, 2023.

¹⁴ Institute for Quality and Efficiency in Health Care (2017) Ribociclib (breast cancer) – Benefit assessment according to Section 35a of the Social Code Book V; Dossier assessment. https://go.sn.pub/kqadrk. Accessed July 11, 2023.

¹⁵ Institute for Quality and Efficiency in Health Care (2020) Apalutamide (prostate cancer) – Benefit assessment according to Section 35a SGB V (expiration of limitation); Dossier assessment. https://go.sn.pub/7dbyze. Accessed July 11, 2023.

¹⁶ Institute for Quality and Efficiency in Health Care (2018) Pertuzumab (breast cancer) – Benefit assessment according to Section 35a of the Social Code Book V; Dossier assessment. https://go.sn.pub/nnvllq. Accessed July 11, 2023.

¹⁷ Institute for Quality and Efficiency in Health Care (2021) Pertuzumab/Trastuzumab (adjuvant breast cancer) – Benefit assessment according to Section 35a of the Social Code Book V; Dossier assessment. https://go.sn.pub/x6kivq. Accessed July 11, 2023.

¹⁸ Institute for Quality and Efficiency in Health Care (2020) Nintedanib (other chronic progressive fibrosing interstitial lung diseases) – Benefit assessment according to Section 35a of the Social Code Book V; Dossier assessment. https://go.sn.pub/4bj058. Accessed July 11, 2023.

¹⁹ Institute for Quality and Efficiency in Health Care (2024) Niraparib/Abiraterone Acetate (Prostate Cancer); Addendum to Project A23-107 (Dossier Assessment). https://go.sn.pub/g13v0o. Accessed May 6, 2024.

²⁰ Institute for Quality and Efficiency in Health Care (2023) Lisocabtagene maraleucel (DLBCL, HGBL, PMBCL, and FL3B, second-line); Benefit assessment according to Section 35a of the Social Code Book V; dossier assessment. https://go.sn.pub/uz4974. Accessed September 16, 2023.

²¹ Institute for Quality and Efficiency in Health Care (2023) Axicabtagene ciloleucel (DLBCL and HGBL, second-line); Benefit assessment according to Section 35a SGB V; dossier assessment. https://go.sn.pub/w6w8p2. Accessed October 10, 2023.

²² Institute for Quality and Efficiency in Health Care (2024) Polatuzumab vedotin (combination with rituximab, cyclophosphamide, doxorubicin, and prednisone; previously untreated DLBCL); Benefit assessment according to Section 35a of the Social Code Book V; dossier assessment. https://go.sn.pub/n4pkoh. Accessed April 2, 2024.

²³ Institute for Quality and Efficiency in Health Care (2024) Nivolumab (NSCLC, neoadjuvant); Addendum to Project A23-74 (dossier assessment). https://go.sn.pub/81yl50. Accessed February 1, 2024.

²⁴ Institute for Quality and Efficiency in Health Care (2022) Pembrolizumab (breast cancer) – Benefit assessment according to Section 35a of the Social Code Book V; Dossier assessment. https://go.sn.pub/6f2qkt. Accessed July 11, 2023.

²⁵ Institute for Quality and Efficiency in Health Care (2025) [A24-61] Scientific report on clinical studies in the therapeutic area of ​​wound treatment. https://go.sn.pub/t18565. Accessed January 7, 2025.

²⁶ Schuster Bruce C, Brhlikova P, Heath J, McGettigan P (2019) The use of validated and nonvalidated surrogate endpoints in two European Medicines Agency expedited approval pathways: A cross-sectional study of products authorized 2011-2018. PLoS Med 16(9):e1002873. doi:10.1371/journal.pmed.1002873.

²⁷ Gyawali B, Hey SP, Kesselheim AS (2020) Evaluating the evidence behind the surrogate measures included in the FDA's table of surrogate endpoints as supporting approval of cancer drugs. EClinicalMedicine 21:100332. doi:10.1016/j.eclinm.2020.100332.

²⁸ Wallach JD, Yoon S, Doernberg H (2024) Associations Between Surrogate Markers and Clinical Outcomes for Nononcologic Chronic Disease Treatments. JAMA 331(19):1646-1654. DOI: 10.1001/Jama.2024.4175.

²⁹ Wieseler B, McGauran N, Kaiser T (2019) New Drugs: Where did we go wrong and what can we do Better? BMJ 366: L4340. DOI: 10.1136/BMJ.L4340.

³⁰ Institute for Quality and Economics in Health Care (2023) Finerenon (renal failure, stage 3 and 4); Addendum to the A23-15 project (dossier evaluation). https://go.sn.pub/qmqozn. Accessed 23.08.2023.

³¹ Thomas Kaiser; Institute for Quality and Economics in Health Care (2016) Clinically relevant study endpoints and surrogate in the early benefit assessment. Interdisciplinary platform for benefit assessment issue 2 January 2016

³² Institute for Quality and Economic Healthcare (2023) Risankizumab (Crohn's disease); Addendum to the project A22-133 (dossier evaluation). https://go.sn.pub/42elj9. Accessed 11.07.2023.

³³ Institute for Quality and Economic Healthcare (2024) Vadadustat in symptomatic anemia due to chronic kidney disease: chance wasted. https://go.sn.pub/a2uk9c.

³⁴ Institute for Quality and Economics in Health Care (2024) Insulin ICODEC (Diabetes Mellitus type 2) - Benefit assessment according to § 35a SGB V. https://go.sn.pub/byfekc.

³⁵ Kramer L, Moos M, Thaa B (2024) Health-Related Quality of Life (HRQOL) in German Early Benefit Assessment: The Importance of Disease-Specific Instruments. Z Evid Continuity Healthy 186: 1-9. DOI: 10.1016/J.ZEFQ.2024.02.003.

³⁶ Institute for Quality and Economics in Health Care (2022) Criticism of study planning: Patient reports often raised far too short. https://go.sn.pub/em0o9l.

³⁷ Institute for Quality and Economic Healthcare (2023) Long -term collection of patient -reported endpoints in oncological studies: important and feasible. https://go.sn.pub/bdb35h.

³⁸ PE M, Alanya A, Falk RS (2023) Setting International Standards in Analyzing Patient-Reported Outcomes and Quality of Life Endpoints in Cancer Clinical Trials Innovative Medicines Initiative (SISAQOL-IMI): StakeHolder views, objectives, and procedures. Lancet Oncol 24 (6): E270-E283. DOI: 10.1016/S1470-2045 (23) 00157-2.

³⁹ Institute for Quality and Economic Healthcare (2021) Clinical relevance of patient -reported endpoints: New threshold proves to be practical. https://go.sn.pub/jmstw4.