Decision thresholds for benefit-risk judgments: the new GRADE Working Group Guidance

Suzumura, Erica Aranha; Jahn, Beate; Siebert, Uwe; Soarez, Patricia Coelho de

doi:10.1590/S2237-96222026v35e20251401.en

One of the publications issued in 2025 by the Grading of Recommendations Assessment, Development and Evaluation (GRADE) Working Group, GRADE Guidance 42, established a formal and transparent methodology for applying decision thresholds to determine the magnitude of the effects of a given intervention or technology on health outcomes¹. Historically, judgments about the magnitude of effect (e.g., trivial, small, moderate or large) have been inconsistent and lacking in explicit justification. Decision thresholds are defined as predefined quantitative reference points adapted to specific outcomes. These thresholds serve as an additional tool among available methodologies² to weigh the balance between desirable effects (benefits) and undesirable effects (risks or harms) in the processes of developing clinical guidelines and in health technology assessment studies.

Selecting outcomes representing benefits and risks

The process begins with the selection and prioritization of health outcomes for which the effects of the use of a health technology will be assessed. Judgments regarding decision thresholds are made at the individual outcome level. Therefore, desirable outcomes (benefits) and undesirable outcomes (risks) most important to the health service user in the context of the benefit-risk assessment in question must be identified, ensuring that each is clearly defined (including how the outcome is measured/assessed and the time to measurement)³. The number of outcomes should be manageable in the context of the assessment (typically up to seven)¹.

Approaches for defining decision thresholds

GRADE suggests two main approaches and a step-by-step process for defining decision thresholds in judgments about the magnitude of benefits and risks. The first approach is based on generic coefficients derived from a methodological study conducted and published by the GRADE Working Group⁴. The second is based on expert opinion. It is important to note that these two approaches can be combined to verify the agreement of the decision thresholds defined through the different approaches¹.

Approach 1: based on generic coefficients

Methodological principle

The methodology for establishing decision thresholds is based on the expected utility theory. Utility values (ranging from 0, for the state of death, to 1, for full health) or its complement, disutility (1 - utility), serve as a basis for weighting the value of outcomes in such a way that they reflect the importance of that outcome in the population of interest⁵^{), (}⁶. Utility values can be elicited through direct methods - such as the standard gamble method, time trade-off, or visual analog scale - or indirect methods - such as generic questionnaires for quality of life assessment (e.g., SF-6D, EQ-5D or HUI-3)⁷.

Identifying utility

Each outcome should be weighted by its respective utility value, which reflects the importance and preference for that outcome⁵^{), (}⁶. The closer the utility value of an outcome is to 0, the more important that outcome is considered in decision-making. Ideally, utilities should be obtained from systematic reviews or primary studies using direct or indirect preference elicitation methods⁷. If neither of these options is feasible, utility determination can be done through consultation with experts, groups responsible for developing guidelines or external stakeholders, such as health service user representatives⁸.

Calculating decision thresholds

The GRADE Working Group uses four categories of effect magnitude: trivial or none, small, moderate and large¹. Three decision thresholds must be determined for each outcome, separating the four magnitude of effect categories. The decision thresholds are expressed as absolute risk difference per 1,000 people.

Threshold calculation is done by applying linear regression equations to the disutility value (or 1 - utility) associated with the outcome. This ensures that the resulting decision thresholds are scaled according to the perceived importance of the specific outcome. The GRADE Working Group provides an online calculator for calculating decision thresholds from utility or disutility values (https://fmup.shinyapps.io/dt_calc)¹. Thresholds can be calculated manually using the following regression equations, which employ generic coefficients.

Decision threshold distinguishing between “trivial or no” effect vs. “small” effect (decision threshold_{trivial/small}):

Decision threshold=0.073-(0.061 x disutility)

Decision threshold distinguishing between “small” effect vs. “moderate” effect (decision threshold_{small/moderate}):

Decision threshold=0.180-(0.146 x disutility)

Decision threshold distinguishing between “moderate” effect vs. “large” effect (decision threshold_{moderate/large}):

Decision threshold=0.338-(0.271 x disutility)

An outcome associated with a high disutility value (e.g., 0.9) will have lower decision thresholds compared to an outcome with low disutility (e.g., 0.2). This implies that, for a more important outcome, a smaller absolute risk difference is needed to categorize the magnitude of effect as “large” than for a less important outcome.

Approach 2: based on expert opinion

This alternative relies on direct consultation with experts, decision-makers or other stakeholders. Experts are asked to explicitly state the absolute risk difference per 1,000 people that, in their judgment, separates the magnitude of effect categories for a specific outcome¹. While simpler to implement, this approach must be carefully managed to mitigate potential cognitive biases and should be carried out without knowledge of the actual magnitude of the intervention effect (meta-analysis result)¹. Because this approach involves direct determination of decision thresholds, obtaining utility values can be ignored when using this approach.

However, the GRADE Working Group encourages adoption of a combined approach, in which decision threshold sets are calculated using the linear regression equations presented above (Approach 1) and obtained through expert opinion (Approach 2). This allows decision-makers to compare decision threshold sets, and any significant discrepancy requires explicit discussion and transparent justification for the final selection of thresholds used in decision-making¹.

Identifying the magnitude of effect

Once the decision thresholds have been defined, the summary effect of the evidence synthesis (i.e., the meta-analysis result) is compared or positioned in relation to these thresholds. This comparison is fundamental for determining and classifying the magnitude of the effect as trivial/none, small, moderate or large. For this assessment to be done correctly, the meta-analysis result must be expressed as the absolute risk difference per 1,000 people. Previous publications have provided instructions for calculating the absolute risk difference from the relative risk or time-to-event measures⁹^{), (}¹⁰.

Two examples of decision thresholds calculated based on generic coefficients using the online calculator (https://fmup.shinyapps.io/dt_calc/) are illustrated in this paper (Figure 1). When assessing the outcome of deep vein thrombosis, the intervention demonstrated an absolute risk difference of -70 events per 1,000 people in relation to the comparator. Based on its associated disutility of 0.42¹¹, the following thresholds were established for categorizing the magnitude of the effect: 47 events (threshold_{trivial/small}), 119 events (threshold_{small/moderate}) and 224 events (threshold_{moderate/large}) per 1,000 people. Considering these limits, the effect estimate fell within the small magnitude benefit range, since the reduction of 70 events per 1,000 people exceeded the trivial effect size threshold but remained below the moderate effect size threshold (Figure 1A).

Figure 1
Decision thresholds (DT) calculated from the disutility values for (A) the deep vein thrombosis outcome and (B) the intracranial hemorrhage outcome, showing the grading of the summary effect estimates, expressed as absolute risk difference (ARD), and their respective 95% confidence intervals (95%CI) in relation to the thresholds that determine the magnitude of the effect (trivial/none, small, moderate or large).

Decision thresholds are presented in this paper for the intracranial hemorrhage outcome, the effect estimate of which demonstrated an absolute risk difference of +70 events per 1,000 people in relation to the comparator (Figure 1B). Considering its associated disutility of 0.88¹¹, the thresholds calculated for categorizing the magnitude of effect were: 19 events (threshold_trivial/smal), 52 events (threshold_{small/moderate}) and 100 events (threshold_{moderate/large}) per 1,000 people. In this scenario, the estimated effect of +70 events per 1,000 people was placed in the moderate magnitude risk range, situated between the small/moderate and moderate/large thresholds.

Considerations

Overall balance and correlated outcomes

Although decision thresholds provide clarity for individual outcomes, overall judgment on the benefit-risk balance continues to be a key challenge, especially when outcomes are correlated (e.g., different manifestations of the same event). The GRADE Working Group recommends selecting independent outcomes or, if correlation is unavoidable, considering the outcome with the highest certainty of evidence to guide the judgment of balance¹².

Incorporation of uncertainty

Although the effect of the technology under evaluation is represented by an effect estimate and its 95% confidence interval (95%CI), the decision thresholds for distinguishing the effect size are point values¹. However, the confidence interval around the effect estimate can be used to judge certainty regarding the magnitude. For example, if the entire 95%CI of the absolute risk difference is in the “small” magnitude category, there is greater certainty in judging the effect size as “small”.

Limitations

Generic coefficients are based on hypothetical scenarios and may not be perfectly generalizable to all clinical contexts. Furthermore, obtaining standardized and robust utility values for all relevant health states continues to be a practical challenge¹. The GRADE Working Group emphasizes that decision thresholds are not a substitute for the judgment of experts, methodologists and decision-makers, but rather an explicit tool to structure, inform discussion and support clinical guideline development processes and health technology assessment studies¹. Therefore, expressing the balance between benefits and risks additionally as an incremental harm-benefit ratio remains important for decision-makers, service providers and service users in shared decision-making processes.

Data Availability

The data are available within the body of the manuscript.

REFERENCES

¹ Wiercioch W, Morgano GP, Piggott T, Nieuwlaat R, Neumann I, Sousa-Pinto B, et al. GRADE Guidance: using thresholds for judgments on health benefits and harms in decision making (GRADE Guidance 42). Ann Intern Med. 2025; 178(11): 1644-1652.
² Suzumura EA, Ascef BO, Maia FHA, Bortoluzzi AFR, Domingues SM, Farias NS, et al. Methodological guidelines and publications of benefit-risk assessment for health technology assessment: a scoping review. BMJ Open. 2024; 14(6): e086603.
³ Wiercioch W, Nieuwlaat R, Zhang Y, Alonso-Coello P, Dahm P, Iorio A, et al. New methods facilitated the process of prioritizing questions and health outcomes in guideline development. J Clin Epidemiol. 2022; 143: 91-104.
⁴ Morgano GP, Wiercioch W, Piovani D, Neumann I, Nieuwlaat R, Piggott T, et al. Defining decision thresholds for judgments on health benefits and harms using the grading of recommendations assessment, development, and evaluation (GRADE) evidence to decision (EtD) frameworks: a randomized methodological study (GRADE-THRESHOLD). J Clin Epidemiol. 2025; 179: 111639.
⁵ Zhang Y, Alonso-Coello P, Guyatt GH, Yepes-Nuñez JJ, Akl EA, Hazlewood G, et al. GRADE Guidelines: 19. Assessing the certainty of evidence in the importance of outcomes or values and preferences-Risk of bias and indirectness. J Clin Epidemiol. 2019; 111: 94-104.
⁶ Zhang Y, Coello PA, Guyatt GH, Yepes-Nuñez JJ, Akl EA, Hazlewood G, et al. GRADE guidelines: 20. Assessing the certainty of evidence in the importance of outcomes or values and preferences-inconsistency, imprecision, and other domains. J Clin Epidemiol. 2019; 111: 83-93.
⁷ Arnold D, Girling A, Stevens A, Lilford R. Comparison of direct and indirect methods of estimating health state utilities for resource allocation: review and empirical analysis. BMJ. 2009; 339: b2688.
⁸ Wiercioch W, Nieuwlaat R, Dahm P, Iorio A, Mustafa RA, Neumann I, et al. Development and application of health outcome descriptors facilitated decision-making in the production of practice guidelines. J Clin Epidemiol. 2021; 138: 115-27.
⁹ Newcombe RG, Bender R. Implementing GRADE: calculating the risk difference from the baseline risk and the relative risk. Evid Based Med. 2014; 19(1): 6-8.
¹⁰ Skoetz N, Goldkuhle M, van Dalen EC, Akl EA, Trivella M, Mustafa RA. GRADE guidelines 27: how to calculate absolute effects for time-to-event outcomes in summary of findings tables and evidence profiles. J Clin Epidemiol. 2020; 118: 124-31.
¹¹ Cuker A, Tseng EK, Schünemann HJ, Angchaisuksiri P, Blair C, Dane K, et al. American Society of Hematology living guidelines on the use of anticoagulation for thromboprophylaxis for patients with COVID-19: March 2022 update on the use of anticoagulation in critically ill patients. Blood Adv. 2022; 6(17): 4975-82.
¹² Balshem H, Helfand M, Schünemann HJ, Oxman AD, Kunz R, Brozek J, et al. GRADE guidelines: 3. Rating the quality of evidence. J Clin Epidemiol. 2011; 64(4): 401-6.

Funding
National Council for Scientific and Technological Development (CNPq, Process 400224/2022-4).
Use of generative artificial intelligence
Not used.

Edited by

Editor-in-Chief:
Jorge Otávio Maia Barreto 0000-0002-7648-0472

Scientific Editor:
Everton Nunes da Silva 0000-0001-8747-4185

Publication Dates

Publication in this collection
20 July 2026
Date of issue
2026

History

Received
05 Jan 2026
Accepted
15 Mar 2026

This is an open-access article distributed under the terms of the Creative Commons Attribution License

[1] ¹ Wiercioch W, Morgano GP, Piggott T, Nieuwlaat R, Neumann I, Sousa-Pinto B, et al. GRADE Guidance: using thresholds for judgments on health benefits and harms in decision making (GRADE Guidance 42). Ann Intern Med. 2025; 178(11): 1644-1652.

[2] ² Suzumura EA, Ascef BO, Maia FHA, Bortoluzzi AFR, Domingues SM, Farias NS, et al. Methodological guidelines and publications of benefit-risk assessment for health technology assessment: a scoping review. BMJ Open. 2024; 14(6): e086603.

[3] ³ Wiercioch W, Nieuwlaat R, Zhang Y, Alonso-Coello P, Dahm P, Iorio A, et al. New methods facilitated the process of prioritizing questions and health outcomes in guideline development. J Clin Epidemiol. 2022; 143: 91-104.

[4] ⁴ Morgano GP, Wiercioch W, Piovani D, Neumann I, Nieuwlaat R, Piggott T, et al. Defining decision thresholds for judgments on health benefits and harms using the grading of recommendations assessment, development, and evaluation (GRADE) evidence to decision (EtD) frameworks: a randomized methodological study (GRADE-THRESHOLD). J Clin Epidemiol. 2025; 179: 111639.

[5] ⁵ Zhang Y, Alonso-Coello P, Guyatt GH, Yepes-Nuñez JJ, Akl EA, Hazlewood G, et al. GRADE Guidelines: 19. Assessing the certainty of evidence in the importance of outcomes or values and preferences-Risk of bias and indirectness. J Clin Epidemiol. 2019; 111: 94-104.

[6] ⁶ Zhang Y, Coello PA, Guyatt GH, Yepes-Nuñez JJ, Akl EA, Hazlewood G, et al. GRADE guidelines: 20. Assessing the certainty of evidence in the importance of outcomes or values and preferences-inconsistency, imprecision, and other domains. J Clin Epidemiol. 2019; 111: 83-93.

[7] ⁷ Arnold D, Girling A, Stevens A, Lilford R. Comparison of direct and indirect methods of estimating health state utilities for resource allocation: review and empirical analysis. BMJ. 2009; 339: b2688.

[8] ⁸ Wiercioch W, Nieuwlaat R, Dahm P, Iorio A, Mustafa RA, Neumann I, et al. Development and application of health outcome descriptors facilitated decision-making in the production of practice guidelines. J Clin Epidemiol. 2021; 138: 115-27.

[9] ⁹ Newcombe RG, Bender R. Implementing GRADE: calculating the risk difference from the baseline risk and the relative risk. Evid Based Med. 2014; 19(1): 6-8.

[10] ¹⁰ Skoetz N, Goldkuhle M, van Dalen EC, Akl EA, Trivella M, Mustafa RA. GRADE guidelines 27: how to calculate absolute effects for time-to-event outcomes in summary of findings tables and evidence profiles. J Clin Epidemiol. 2020; 118: 124-31.

[11] ¹¹ Cuker A, Tseng EK, Schünemann HJ, Angchaisuksiri P, Blair C, Dane K, et al. American Society of Hematology living guidelines on the use of anticoagulation for thromboprophylaxis for patients with COVID-19: March 2022 update on the use of anticoagulation in critically ill patients. Blood Adv. 2022; 6(17): 4975-82.

[12] ¹² Balshem H, Helfand M, Schünemann HJ, Oxman AD, Kunz R, Brozek J, et al. GRADE guidelines: 3. Rating the quality of evidence. J Clin Epidemiol. 2011; 64(4): 401-6.