Free scoping read. A written assessment of your manuscript in two business days, at no cost.

Start one
Contents

Academy · Research Data, Analytics & AI · Decision Analysis Primer

A Primer on Decision Analysis for Clinical Investigators: Decision Trees, Sensitivity Analysis and Patient Preferences

Decision trees and sensitivity analysis provide clinical investigators with a structured method for weighing treatment choices under uncertainty, rather than relying on intuition alone.

Pubrica AcademyWritten by a clinical epidemiologist Reviewed bya senior biostatistician Published10 October 2026 Last checked10 October 2026 Reading time7 minutes Sources10 peer-reviewed

Most clinical decisions are made under incomplete information: a treatment’s true benefit, its risk of harm, and a patient’s own preferences are all estimates, not certainties. Clinical decision analysis is the branch of biostatistics designed specifically to address this uncertainty: it formalizes a choice into a structure that can be examined, challenged, and tested for robustness, rather than debated informally.

This primer focuses on two of its core tools for clinical investigators and decision-makers: decision trees, which set out the alternatives and their probable outcomes, and sensitivity analysis, which tests how much a recommendation depends on uncertain inputs.

Learning objectives

  • What a clinical decision tree represents and how it differs from a flowchart
  • How probabilities and utilities combine inside a decision tree to produce an expected value
  • What sensitivity analysis adds once a base-case decision tree has been built
  • How these tools incorporate patient preferences alongside clinical evidence

Common misconceptions

  • “A decision tree is just a flowchart of clinical options.” A decision tree assigns probabilities and utilities (quantified values of outcomes) to each branch so the alternatives can be compared on expected value, not merely laid out sequentially.
  • “Sensitivity analysis is an optional add-on once the 'real' analysis is done.” Sensitivity analysis is treated as a core step of decision analysis in the literature, because it directly tests how much the preferred option changes as uncertain probabilities or utilities are varied, which is often the most clinically useful output.
  • “Decision analysis replaces clinical judgment with a formula.” The method is explicitly built to incorporate patient preferences and clinical judgment as inputs (utilities), not to replace them; the formal structure exists to make those judgments explicit and testable, not to eliminate them.

Purpose of clinical decision analysis

Clinical decision analysis applies explicit, quantitative methods to clinical decisions made under uncertainty1. Rather than choosing between treatments by informal weighing of advantages and disadvantages, it builds a structured model of the alternatives, the probability of each possible outcome, and the desirability of each outcome to the people affected by it1.

Structure of a decision tree

A decision tree lays out the available alternatives and the outcomes linked to each one1. At each branch point, a probability is attached to each possible outcome, and a utility — a numerical value representing how good or bad that outcome is — is attached to the endpoint1. Multiplying probabilities by utilities along each branch and summing them yields an expected value for each alternative, which allows the investigator to compare options on a common scale rather than by intuition1.

Contribution of sensitivity analysis

Because the probabilities and utilities that feed a decision tree are themselves estimates, often drawn from imperfect literature, a single "base case" result can be misleading on its own. Sensitivity analysis addresses this directly: it lets the analyst test how uncertainty in the underlying data affects the expected clinical outcome, and identify which input variables most influence the final recommendation1. In practice, this means varying one input at a time (or several together) across a plausible range and observing whether the preferred option changes; a result that is robust across that range is considerably more trustworthy than one that reverses with small changes in a single uncertain estimate.

Incorporation of patient preferences

Because utilities represent how desirable an outcome is, decision analysis gives a formal route for patient preference to enter a clinical decision alongside the probability evidence, rather than being treated as a separate, informal conversation1. Two patients facing the same probabilities of benefit and harm from a treatment can rationally reach different decisions if their utilities for those outcomes differ — and the decision tree makes that difference visible and explicit rather than implicit.

Limitations of the method

Decision analysis is only as reliable as the probability and utility estimates that inform it. When good evidence for a given probability does not exist, analysts substitute expert judgment or a plausible range tested through sensitivity analysis; this is a limitation to be disclosed rather than concealed. For clinical investigators, the practical value of the method lies less in producing a single “correct” number than in making every assumption behind a treatment recommendation explicit, where it can be checked.

Markov models as an extension of the decision tree

A standard decision tree works well for a one-time choice with a short time horizon, but many clinical decisions — whether to start a chronic therapy, how to manage a relapsing condition — involve a patient moving between health states over years, sometimes returning to a state they already occupied. A simple tree cannot represent that kind of recurring, open-ended process without becoming unmanageably large. Sonnenberg and Beck's widely used 1993 methodological guide describes the Markov model as the standard solution: the clinical course is represented as a set of mutually exclusive health states (e.g., well, relapsed, dead), together with the probabilities of transitioning between them over a fixed cycle length (for example, one month), run forward for as many cycles as the clinical question requires2. Because the model can cycle indefinitely, it can represent lifetime risk, recurrence, and long-run cost or survival in a way a single-pass decision tree cannot, at the cost of requiring a full transition-probability matrix rather than a small number of branch probabilities2. A widely used textbook treatment of these methods extends the Markov approach into full cost-effectiveness modeling practice, covering model structure, parameter estimation, and reporting conventions in detail10.

Sources of utility values

A decision tree's utility values are not arbitrary — they are usually drawn from a specific, named elicitation method, most commonly the standard gamble or the time trade-off technique. In the foundational 1978 study introducing this approach to health-state valuation, Sackett and Torrance asked members of the general public to value chronic health states using the time trade-off method, in which a respondent states how much remaining lifespan in a health state they would be willing to give up in exchange for that time being spent in full health3. The resulting utility, scored on a scale typically anchored at 0 (equivalent to death) and 1 (full health), becomes the numeric input a decision tree multiplies against outcome probabilities. Because different elicitation methods (standard gamble, time trade-off, rating scale) and different respondent populations (patients living with a condition versus members of the general public asked to imagine it) can produce materially different utility values for the same health state, a rigorous decision analysis should state explicitly which method and which population generated its utility inputs, rather than reporting a single number as if it were an objective fact.

When a decision tree's utilities are expressed as quality-adjusted life years (QALYs) — survival time weighted by the utility of the health state lived in — the same tree structure extends naturally into a formal cost-effectiveness analysis, comparing one strategy's extra cost against its extra QALYs relative to another. The methodological standard most frequently cited for how to conduct this kind of analysis is the 1996 report of the U.S. Panel on Cost-Effectiveness in Health and Medicine, published in JAMA, which recommended a standardized "reference case" analysis — specifying a consistent perspective, discount rate, and outcome measure — specifically so that cost-effectiveness results from different studies and different interventions could be meaningfully compared against one another4. Health technology assessment bodies such as the UK's National Institute for Health and Care Excellence have since built their own formal reference case requirements directly on this foundation, specifying exactly how QALYs, costs, and discounting should be handled in decision models submitted for national funding decisions5.

Assessment of whether further research justifies its cost

Sensitivity analysis indicates which uncertain inputs matter most to a decision’s conclusion, but it does not by itself establish whether resolving that uncertainty through further research justifies the expense. A newer analytic extension, the expected value of perfect information (EVPI), assigns a monetary value to eliminating, in principle, all uncertainty in a decision model’s parameters, effectively an upper bound on what a funder should be willing to pay for additional research before a decision is made6. This closes a conceptual loop that traces back to the earliest clinical decision-analysis literature: Pauker and Kassirer's 1980 "threshold approach to clinical decision making" formalized the idea that a rational clinician should order a test only when its result could plausibly cross a probability threshold that would change the treatment decision7 — EVPI is the modern, quantified descendant of that same basic question applied to research investment rather than a single diagnostic test.

Importance of model documentation alongside mathematical structure

A 1987 progress report in Annals of Internal Medicine by Kassirer, Moskowitz, Lau, and Pauker, reviewing the first decade of clinical decision analysis in practice, observed that the method’s credibility depends heavily on investigators reporting their complete reasoning (every probability source, every utility elicitation method, every structural assumption) rather than presenting only a final recommendation8. That principle was later formalized at a much larger scale by the ISPOR-SMDM Modeling Good Research Practices Task Force, whose 2012 overview report sets out a shared checklist for structuring, documenting, and validating decision-analytic models so that an independent reviewer can evaluate whether a model’s conclusions follow from its stated assumptions, rather than accepting the output without scrutiny9.

Key points

Key points

Decision trees convert an informal clinical judgement into an explicit, verifiable model, and sensitivity analysis indicates the extent to which the conclusion of that model depends on its least certain assumptions.

Practical application

Practical application

  1. All clinically plausible alternatives should be specified before values are assigned. Every clinically plausible option and outcome should be set out first; a decision tree built on an incomplete set of branches will mislead regardless of the care taken with the probabilities.
  2. Probabilities should be derived from the literature rather than from intuition. Published incidence, response or complication rates should be used as branch probabilities wherever available, and any branch for which no such evidence exists should be identified.
  3. A sensitivity analysis should be conducted on the most uncertain inputs. The one or two least certain inputs should be identified and varied across a plausible range to determine whether the preferred option changes.
  4. Patient preferences should be represented explicitly through utilities. Where the desirability of an outcome depends on the patient’s own values (for example, tolerance of a side effect relative to residual symptoms), this should be represented in the utility value rather than decided informally.

Frequently asked questions

No. A flowchart sequences steps; a decision tree attaches probabilities and utilities to branches so alternatives can be compared on expected value.

There is no fixed rule, but the primer literature frames the purpose as identifying which variables most affect the final outcome, so a reasonable approach is to begin with the one or two least certain inputs and to expand the analysis if the conclusion proves fragile.

It can, using expert-elicited or literature-range estimates, but any such substitution should be explicitly tested with sensitivity analysis rather than presented as if it were a firm outcome rate.

References

  1. Aleem I, Jalal H, Aleem IS, Sheikh AA, Bhandari M. Clinical decision analysis: incorporating the evidence with patient preferences. Patient Preference and Adherence. 2008;3:21-24. read 10 October 2026. View source
  2. Sonnenberg FA, Beck JR. Markov models in medical decision making: a practical guide. Medical Decision Making. 1993;13(4):322-338. read 10 October 2026. View source
  3. Sackett DL, Torrance GW. The utility of different health states as perceived by the general public. Journal of Chronic Diseases. 1978;31(11):697-704. read 10 October 2026. View source
  4. Weinstein MC, Siegel JE, Gold MR, Kamlet MS, Russell LB. Recommendations of the Panel on Cost-Effectiveness in Health and Medicine. JAMA. 1996;276(15):1253-1258. read 10 October 2026. View on PubMed
  5. National Institute for Health and Care Excellence. NICE Health Technology Evaluations: The Manual. read 10 October 2026. View source
  6. Heath A, Manolopoulou I, Baio G. A review of methods for the analysis of the expected value of information. arXiv:1507.02513. 2015. read 10 October 2026. View source
  7. Pauker SG, Kassirer JP. The threshold approach to clinical decision making. New England Journal of Medicine. 1980;302(20):1109-1117. read 10 October 2026. View source
  8. Kassirer JP, Moskowitz AJ, Lau J, Pauker SG. Decision analysis: a progress report. Annals of Internal Medicine. 1987;106(2):275-291. read 10 October 2026. View source
  9. Caro JJ, Briggs AH, Siebert U, Kuntz KM; ISPOR-SMDM Modeling Good Research Practices Task Force. Modeling good research practices—overview: a report of the ISPOR-SMDM Modeling Good Research Practices Task Force-1. Value in Health. 2012;15(6):796-803. read 10 October 2026. View source
  10. Briggs A, Claxton K, Sculpher M. Decision Modelling for Health Economic Evaluation. Oxford University Press; 2006. read 10 October 2026. View source

Every source above is a peer-reviewed article, a publisher’s record of one, or a standard issued by the body that owns it. Each carries the date it was last read. Where a standard moves, this article is revised rather than left up with a note.

How to cite this article

Pick your referencing style and copy it. Your university or target journal may vary the punctuation — check its guide if it has one.

Pubrica Academy. (2026, October 10). A Primer on Decision Analysis for Clinical Investigators: Decision Trees, Sensitivity Analysis and Patient Preferences. Pubrica. https://pubrica.com/academy/concepts-definitions/a-primer-for-clinical-investigators-and-decision-makers-in-clinical-epidemiology-and-biostatistics/

This article is written to be learned from. Cite it if you are referring to what it says; if you are using the evidence behind it, cite the sources listed above instead.

Decision-tree build checklist

This checklist should be completed before a decision-analysis recommendation is presented to clinical stakeholders.

[ ] All clinically realistic alternatives are represented as branches
[ ] Each outcome has a sourced probability, not an assumed one
[ ] Each outcome has a utility reflecting patient-relevant desirability
[ ] A base-case expected value has been calculated for each alternative
[ ] Sensitivity analysis has been run on the least-certain inputs
[ ] The recommendation's robustness (or fragility) across that range is reported

Pubrica AcademyWritten and maintained by Pubrica’s subject matter experts. Articles are revised when the underlying standards change, and each article states the date on which it was last reviewed.

How Pubrica can help

Pubrica's biostatistics team builds and sensitivity-tests decision-analytic models that make the probability and preference assumptions behind a treatment recommendation explicit and defensible.

Find something else

Search the Pubrica Academy

Access is free and requires no account. Enter a topic, or open the full library to filter by category, sub-topic or resource type.

Sensitivity-tested decision models for treatment decisions

A scoping assessment is free and takes two business days. It provides a written assessment of what the work requires and which specialist would undertake it.

Get a free quote

Regulations

Check your project against this rule

  1. 1Your work
  2. 2Your details
  3. 3Send
What do you need?

Material you send is held under signed confidentiality, with access limited to the assigned team and encrypted transfer. It is not used to train anything and is not passed to third-party tools. Ask for the agreement before you send anything you would rather cover first.