Clinical decision-making: evidence-based care

In a systematic review with 48 studies and 13,011 professionals, doctors rarely got it right about how much benefit and harm a treatment, a test or a screening brings (Hoffmann and Del Mar, 2017). In comparisons with the correct answer about benefits, most got it right in only 3 of 28 outcomes, or 11% (Hoffmann and Del Mar, 2017). Before knowing how to interpret a result, you need to decide whether that test or that approach is worth it, and that decision depends on estimating gains and losses.

NEXO is a daily medical diagnosis game: one clinical case a day, six progressive clues, and ICD-10 proximity feedback across five levels.

This text deals with clinical decision-making from the point of view of someone who studies: when and why to order a test or choose an approach. It goes through the expected benefits and harms, the strength of a recommendation and the level of evidence behind it, low-value tests and approaches, and shared decision-making with the patient. The interpretation of the result, with pre-test and post-test probability, has its own text about pre-test probability.

Key points

• Deciding on a test or an approach starts with a question: does the result or the treatment change something that matters for this patient?
• Patients and doctors tend to overestimate benefits and underestimate harms, according to two systematic reviews (Hoffmann and Del Mar, 2015; Hoffmann and Del Mar, 2017).
• The GRADE system separates two things: the quality of the evidence and the strength of the recommendation (Guyatt et al., 2008).
• In the cardiology guidelines studied, most recommendations come from lower-level evidence or expert opinion (Tricoci et al., 2009; Fanaroff et al., 2019).
• Low-value tests and approaches are common and can cause physical and psychological harm (Brownlee et al., 2017).
• Shared decision-making support tools increase patients' knowledge and their participation in the choice (Stacey et al., 2024).

Summary

• Before ordering: the question that decides
• Benefits and harms: what we expect and what happens
• Guidelines, strength of the recommendation and level of evidence
• Low-value tests and approaches
• Shared decision-making with the patient
• How to train clinical decision-making
• Frequently asked questions
• Clinical decision checklist

Before ordering: the question that decides

Clinical decision-making is choosing, with a patient in front of you, what to do next: order a test, start a treatment, observe or refer. Every one of these choices has three ingredients: what is known about the patient, what the evidence says about the options and what matters to the person being cared for.

Hypothesis before the order

Every test answers a question, and the question comes from the hypothesis. Whoever orders a test without knowing which hypothesis it tests tends to receive a result without knowing what to do with it. That is why the decision starts in diagnostic reasoning: list the hypotheses, as in the text about differential diagnosis, and prioritize by probability and severity, as in the text about clinical reasoning.

The thresholds, in one sentence

The threshold approach says that a test is only worth it in an intermediate range of probability, between the point where you neither test nor treat and the point where you treat without testing. The full explanation, with a published example, is in the section about the two thresholds. Here, the focus is what comes along with that calculation: the harms, the strength of the evidence and the patient's preference.

Four questions before deciding

As a practical suggestion for the rotation, ask these four questions before ordering a test or proposing an approach:

1. What question does this answer? Which hypothesis the test checks, or which problem the approach treats.
2. What changes if it goes right or wrong? If the positive and the negative result lead to the same approach, the test adds little to the decision. The version with probability is in the questions before ordering.
3. What harm can it cause? Discomfort, risk of the procedure, incidental findings that lead to new tests, anxiety.
4. What does the patient want? A choice among reasonable options also depends on the values of the person who will live with it.

Editorial example; the decision about each test and each approach is always up to the doctor responsible for the patient.

A teaching example

Imagine an imaging exam ordered "just to be safe", for a common complaint with no alarm sign and with a benign hypothesis well supported by the history and physical exam. The four questions change the conversation with the preceptor: which hypothesis the image would test, what would change with a normal or altered result, what an incidental finding would bring, and what the patient expects from the visit. The right answer depends on the case, the service's guideline and the preceptor; the gain is arriving at the discussion with the reason for the order said out loud.

Generic teaching example; it does not apply to any specific complaint or exam.

Benefits and harms: what we expect and what happens

Deciding well requires estimating the size of the benefit and the harm of each option. Two systematic reviews by the same group show that this estimate often comes out skewed, both on the patient side and on the physician side.

What patients expect

A systematic review gathered 35 studies with 27,323 patients who estimated benefits or harms of treatments, tests and screenings (Hoffmann and Del Mar, 2015). Among the 34 benefit outcomes with overestimation data, most participants overestimated the benefit in 22, or 65% (Hoffmann and Del Mar, 2015). Among the 15 harm outcomes with underestimation data, most underestimated the harm in 10, or 67% (Hoffmann and Del Mar, 2015).

The authors conclude that physicians should discuss accurate and balanced information about benefits and harms with patients, so that they form realistic expectations and decide informed (Hoffmann and Del Mar, 2015).

What physicians expect

The review on professionals included 48 studies, 20 on treatment, 20 on imaging exams and 8 on screening (Hoffmann and Del Mar, 2017). In the harm outcomes compared with the correct answer, most got it right in 9 of 69, or 13% (Hoffmann and Del Mar, 2017). The errors went both ways, but physicians underestimated harms more often than they overestimated them and overestimated benefits more often than they underestimated them (Hoffmann and Del Mar, 2017).

Who estimated	Studies and participants	Main finding
Patients	35 studies and 27,323 people (Hoffmann and Del Mar, 2015)	Most overestimated benefits in 65% of the outcomes with data (Hoffmann and Del Mar, 2015)
Patients	15 harm outcomes with data (Hoffmann and Del Mar, 2015)	Most underestimated harms in 67% of these outcomes (Hoffmann and Del Mar, 2015)
Physicians	48 studies and 13,011 professionals (Hoffmann and Del Mar, 2017)	Most got the benefit right in 11% of the compared outcomes (Hoffmann and Del Mar, 2017)
Physicians	69 compared harm outcomes (Hoffmann and Del Mar, 2017)	Most got the harm right in 13% of these outcomes (Hoffmann and Del Mar, 2017)

What this changes for those who study

The authors point out that inaccurate perceptions about benefits and harms tend to lead to worse management choices (Hoffmann and Del Mar, 2017). For the student, the practical lesson is to treat the size of the effect as part of the content: when studying a treatment or a test, note how much it helps, for whom and with what harm, and where that number comes from.

Guidelines, strength of recommendation and level of evidence

Guidelines gather the evidence and translate the result into recommendations. Knowing how to read a recommendation, and not just follow it, is part of clinical decision-making. How to study protocols in the daily routine of the internship is in the text on how to study during the internship; here the subject is what the classification of a recommendation means.

Quality of evidence and strength of recommendation

GRADE was born because guidelines were inconsistent in how they classified the quality of evidence and the strength of recommendations (Guyatt et al., 2008). The system classifies the quality of evidence into four categories, high, moderate, low and very low, according to confidence that the effect estimate is correct (Guyatt et al., 2008). For strength, it offers two degrees, strong and weak, and panels may prefer terms such as "conditional" instead of weak (Guyatt et al., 2008).

What makes a recommendation strong or weak

When the desirable effects clearly outweigh the undesirable ones, or clearly do not outweigh them, a strong recommendation is warranted (Guyatt et al., 2008). Uncertainty about this balance, due to low-quality evidence or to very evenly matched effects, leads to a weak recommendation (Guyatt et al., 2008). The authors recall that, when the balance is close or uncertain, the physician needs to help the patient weigh the effects according to the patient's values and preferences (Guyatt et al., 2008).

When reading a recommendation	What to ask	What changes in the decision
Strong strength	Is the balance between benefit and harm clear?	It tends to apply to most patients
Weak or conditional strength	What makes the balance uncertain?	It opens room for the patient's preference and for shared decision-making
High or moderate evidence	What studies does it come from?	More confidence that the estimated effect is close to the true one
Low or very low evidence	Is there expert opinion in place of data?	More chance the recommendation will change with new studies

The "what to ask" column and the practical reading are a suggestion from the Editorial, based on the GRADE definitions.

How much of the guidelines comes from clinical trials

A study of 53 American cardiology guidelines, with 7,196 recommendations, showed that only a median of 11% of the recommendations in the current guidelines had level of evidence A, while 48% had level C (Tricoci et al., 2009). Even among class I recommendations, the firmest ones, the median with level A was 19% (Tricoci et al., 2009).

A decade later, the analysis of American and European cardiology guidelines found 8.5% of American recommendations and 14.2% of European ones with level A (Fanaroff et al., 2019). In this study, level A means data from several randomized trials or from one large trial, and level C means expert opinion only (Fanaroff et al., 2019). The authors conclude that this pattern did not improve in a relevant way between 2008 and 2018 (Fanaroff et al., 2019).

How to use this during clinical training

These studies deal with cardiology, and other areas may have a different profile. The lesson for those who study is the same: a recommendation has a strength and a basis, and both appear next to the text in most guidelines. When studying a protocol, note the class and the level of each step; when the level is low, the decision calls for more attention to the context and to the patient. If you are interested in how to measure the strength of evidence in medical education, NEXO Researcher brings together those who research learning and reasoning.

Low-value tests and practices

A low-value test or practice is one that brings little or no clinical benefit to the patient in question. The review published in the Lancet series on the topic defines overuse as the provision of medical services with more chance of causing harm than benefit (Brownlee et al., 2017).

How often this happens

The authors describe overuse as well documented in high-income countries, across a wide variety of services, and increasingly recognized in low-income countries (Brownlee et al., 2017). They also point out that measuring overuse directly is difficult, because defining adequate care depends on each patient's preferences and needs (Brownlee et al., 2017).

A study with 1,360,908 Medicare beneficiaries, the American public program for older adults, created 26 measures of low-value services from evidence-based lists (Schwartz et al., 2014). Depending on the measure used, more sensitive or more specific, 25% to 42% of beneficiaries received at least one of these services in 2009 (Schwartz et al., 2014). For laboratory tests, the meta-analysis on overordering and underordering is in the section on the search for certainty.

Where low value appears

The measures by Schwartz and colleagues were grouped into six categories (Schwartz et al., 2014):

Study category	Useful question before ordering
Low-value cancer screening	Does screening make sense for this patient's age and situation?
Low-value diagnostic and preventive tests	What hypothesis does this test check?
Low-value preoperative tests	Does the result change the surgery or the care around it?
Low-value imaging tests	What does the image add to the history and physical examination?
Low-value cardiovascular tests and procedures	Is there a clear indication in the guideline for this case?
Other low-value surgical procedures	What less invasive alternatives were considered?

The categories are from Schwartz and colleagues (2014), in a study with older American adults; the question column is a suggestion from the Editorial and does not replace the service's guideline.

Why low value causes harm

The use of unnecessary services can cause physical and psychological harm to patients and divert resources from health systems (Brownlee et al., 2017). As a teaching example, a test without a clear indication can bring an incidental finding, which calls for another test, which can lead to a procedure, each step with its own risk. The authors also note that the harms of overuse have still been poorly quantified (Brownlee et al., 2017).

Low-value lists

Some medical societies publish lists of low-value tests and practices in specific situations, such as those of the Choosing Wisely campaign. These lists are good study material, as long as they are read at the original source, with the clinical situation to which each item applies. A list item applies to a defined scenario, not as a general prohibition.

Shared decision making with the patient

When there is more than one reasonable option, the best choice also depends on what matters to the patient. Shared decision making is the process in which doctor and patient look together at the options, the benefits, the harms and the preferences before choosing.

The three-talk model

Elwyn and colleagues proposed a practical model of shared decision making in three steps: introducing the choice, describing the options and helping the patient explore preferences and decide (Elwyn et al., 2012). In English, the steps are called choice talk, option talk and decision talk (Elwyn et al., 2012). The model depends on a good relationship in the consultation, so that information is shared and the patient can deliberate and say what they prefer (Elwyn et al., 2012).

Step	What happens	Example of speech (suggestion from the Editorial)
Choice talk	The doctor makes it clear that there is a decision to make (Elwyn et al., 2012)	"For this, there is more than one reasonable path, and your opinion counts."
Option talk	The options are described, often with a support tool (Elwyn et al., 2012)	"I will explain what each path brings that is good and bad."
Decision talk	The patient explores preferences and decides (Elwyn et al., 2012)	"Thinking about what is most important to you, which seems better?"

What the support tools show

The 2024 Cochrane review on decision support tools gathered 209 trials with 107,698 adult participants facing treatment or screening decisions (Stacey et al., 2024). Compared with usual care, the tools increased participants' knowledge and the accuracy of risk perception, with high-certainty evidence (Stacey et al., 2024). They probably increased the consistency between the values reported and the choice made, with moderate-certainty evidence (Stacey et al., 2024).

The review found no difference in decision regret between the groups (Stacey et al., 2024). When the tool was used during the consultation, the visit lasted 1.5 minutes longer; when it was used beforehand, there was no difference in duration (Stacey et al., 2024). The proportion of patients who remained passive in the decision also fell with the tools (Stacey et al., 2024).

Shared decision making in the clerkship

In the clerkship, the final decision belongs to the attending physician, but the student takes part in the conversation. As a practical suggestion, when following a decision with more than one reasonable option, observe whether the three steps appear and note how the preceptor presented the benefits and harms. Then practice the same conversation with colleagues, in a role-play format, as described in the text on simulation of a consultation.

Numbers the patient understands

The reviews by Hoffmann and Del Mar show that both patients and doctors misjudge benefits and harms (Hoffmann and Del Mar, 2015; Hoffmann and Del Mar, 2017). As a practical suggestion, in the conversation, talk about absolute numbers, in frequencies, and show the benefit and the harm side by side. The text about pretest probability explains why natural frequencies are usually easier to understand.

How to train clinical decision making

One way to train clinical decision making is to decide and receive feedback on the choice. Each training format covers one part of it.

In the clerkship, with the preceptor

A good moment is the case presentation, when the student proposes a course of action and the preceptor comments. The models for leading this conversation, such as SNAPPS and the one-minute preceptor, are in the text about clinical reasoning in the clerkship. Bring along the four questions from this text: the reason for the test, what it changes, the possible harm and what the patient wants.

Training the decision in three steps

As a practical suggestion for studying alone or in pairs:

1. Pick a course of action from the day. A test ordered or a treatment started in a patient you followed.
2. Find the recommendation in the source. In the service or society guideline, look for the strength of the recommendation and the level of evidence.
3. Write the balance. In three lines: the expected benefit, the possible harm and the patient's preference, each with its source.

With management cases and simulation

Cases in which the student chooses tests and courses of action and receives feedback on the choice train this step directly; the types of game and what each one trains are in the text about clinical reasoning games. The conversation with the patient about the options is well trained with a simulated patient or colleagues, with a debriefing at the end.

Where NEXO comes in

The decision about tests and management is trained in the clerkship, with management cases and in simulation. NEXO covers the step that comes before: the hypothesis. Each case arrives in up to six clues, one at a time, and each guess shows the proximity in ICD-10, which trains linking the data to a list of hypotheses that adjusts to each clue.

The Daily brings one case per day, the same for everyone; the Archive gathers the previous cases, and Multimedia brings cases with image, audio and other tests. As a practical suggestion, after getting the Daily case right, ask yourself the four questions from this text: which test would confirm the hypothesis, what it would change, what harm it could cause and which course of action the guideline recommends, with the strength and the level of the recommendation. The streak, the ranking and the achievements follow consistency, and the Anki button takes to a card what was missing, as described in the guide on spaced repetition.

To research, bring to the class and create

If you research clinical decision making and medical education and want to propose studies with NEXO, meet NEXO Researcher. To bring NEXO to the medical school, the student league or the residency, there is NEXO Closer. Doctors who review the cases in their specialty take part through NEXO Curator, and those who create health content on social media can take part through NEXO Creators.

Frequently asked questions

What is clinical decision making?

It is choosing, before a patient, what to do next: order a test, treat, observe or refer. The choice brings together three things: the patient's data, the evidence about the options and what matters to the person being cared for.

When is it worth ordering a test?

When the result can change the course of action and the expected benefit outweighs the possible harm. The version with probabilities, with the test and treatment thresholds, is in the text about pretest probability.

What is the GRADE system?

It is a system for grading the quality of evidence, as high, moderate, low and very low, and the strength of recommendations, as strong and weak (Guyatt et al., 2008). It separates how much one trusts the estimated effect from how much one recommends the course of action.

What is the difference between strength of recommendation and level of evidence?

The level of evidence says how much one trusts the studies behind the recommendation. The strength says how clear the balance between benefit and harm is; a weak recommendation indicates that this balance is uncertain or even (Guyatt et al., 2008).

Are guidelines based on clinical trials?

Partly. In the cardiology guidelines studied by Fanaroff and colleagues, 8.5% of American recommendations and 14.2% of European ones had level A, supported by randomized trials (Fanaroff et al., 2019). The rest came from observational studies, a single trial or only expert opinion (Fanaroff et al., 2019).

What is a low-value test?

It is a test that, in the situation at hand, brings little or no benefit and can cause harm. The Lancet review defines overuse as the offering of services with a greater chance of causing harm than benefit (Brownlee et al., 2017).

Does ordering more tests make the diagnosis safer?

It depends on each test. Unnecessary services can cause physical and psychological harm (Brownlee et al., 2017), and an order without a clear hypothesis can bring incidental findings and further tests. As editorial guidance, safety comes from ordering the right test for the right question.

What is shared decision making?

It is the process in which doctor and patient look together at the options, the benefits and the harms and choose taking the patient's preferences into account. The Elwyn model has three steps: present the choice, describe the options and help to decide (Elwyn et al., 2012).

Do decision support tools work?

In the 2024 Cochrane review, they increased patients' knowledge and the accuracy of their risk perception, with high-certainty evidence, with no difference in regret about the decision (Stacey et al., 2024).

How does NEXO help in clinical decision making?

NEXO trains the hypothesis, the step that comes before ordering tests: each case arrives in clues and each guess shows the proximity in ICD-10. The choice of tests and management is trained in the internship, in management cases and in the simulation, and the four questions in this text link one step to the other.

Clinical decision checklist

1. Do I know which hypothesis this test is testing or which problem this management treats?
2. Do I know what changes if the result comes back positive or negative?
3. Have I thought about the possible harm, including incidental findings and further tests?
4. Have I found the recommendation in the source, in the service or society guideline?
5. Have I noted the strength and the level of the recommendation?
6. Have I checked whether the test appears on a low-value list for this situation?
7. Have I estimated benefit and harm in numbers, with the source of each?
8. Have I seen whether there is more than one reasonable option and whether shared decision making fits?
9. Have I explained the options in frequencies, with benefit and harm side by side?
10. Have I taken the reason for the order to the preceptor and recorded the feedback?

References

1. Hoffmann TC, Del Mar C. Clinicians' expectations of the benefits and harms of treatments, screening, and tests: a systematic review. JAMA Intern Med. 2017. https://pubmed.ncbi.nlm.nih.gov/28097303/
2. Hoffmann TC, Del Mar C. Patients' expectations of the benefits and harms of treatments, screening, and tests: a systematic review. JAMA Intern Med. 2015. https://pubmed.ncbi.nlm.nih.gov/25531451/
3. Guyatt GH, Oxman AD, Vist GE, et al. GRADE: an emerging consensus on rating quality of evidence and strength of recommendations. BMJ. 2008. https://pubmed.ncbi.nlm.nih.gov/18436948/
4. Tricoci P, Allen JM, Kramer JM, Califf RM, Smith SC Jr. Scientific evidence underlying the ACC/AHA clinical practice guidelines. JAMA. 2009. https://pubmed.ncbi.nlm.nih.gov/19244190/
5. Fanaroff AC, Califf RM, Windecker S, Smith SC Jr, Lopes RD. Levels of evidence supporting American College of Cardiology/American Heart Association and European Society of Cardiology guidelines, 2008-2018. JAMA. 2019. https://pubmed.ncbi.nlm.nih.gov/30874755/
6. Brownlee S, Chalkidou K, Doust J, et al. Evidence for overuse of medical services around the world. Lancet. 2017. https://pubmed.ncbi.nlm.nih.gov/28077234/
7. Schwartz AL, Landon BE, Elshaug AG, Chernew ME, McWilliams JM. Measuring low-value care in Medicare. JAMA Intern Med. 2014. https://pubmed.ncbi.nlm.nih.gov/24819824/
8. Elwyn G, Frosch D, Thomson R, et al. Shared decision making: a model for clinical practice. J Gen Intern Med. 2012. https://pubmed.ncbi.nlm.nih.gov/22618581/
9. Stacey D, Lewis KB, Smith M, et al. Decision aids for people facing health treatment or screening decisions. Cochrane Database Syst Rev. 2024. https://pubmed.ncbi.nlm.nih.gov/38284415/