Prognostic reasoning: what comes after diagnosis

The resident finalizes the diagnosis, writes the code, and the patient's daughter asks in the hallway: "So, doctor, how long does he have?". The diagnosis answered what the patient has. The daughter's question is another one: what is going to happen. That second question calls for its own reasoning, which medical school usually covers only in passing: prognostic reasoning.

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

This guide shows what prognostic reasoning is, why it has stayed in the shadow of diagnosis, what studies say about the accuracy of medical predictions, and how to train this step starting in medical school.

Key points

• Prognostic reasoning is estimating the likely course of a condition after the diagnosis and stating that estimate with the uncertainty it carries.
• An analysis of editions of a classic textbook shows that prognosis lost ground as diagnosis and treatment gained weight.
• In a study with physicians who referred patients to palliative care, most survival predictions were too optimistic.
• Predictions miss the size but follow the order: the patient the physician thinks will live less does tend to live less.
• Stating the estimate as a probability and in scenarios helps communicate what is known and what is not.

What prognostic reasoning is

Three questions, three lines of reasoning

Faced with a patient, the physician answers three chained questions. The first is diagnostic: what does the patient have? The second is therapeutic: what should be done? The third is prognostic: what is going to happen, with and without what is done? The three support one another, but each has its own logic.

Diagnostic reasoning works with hypotheses and with the way each piece of data changes the list. The guide on clinical reasoning in the clinical cycle shows this path. Therapeutic reasoning weighs benefits and harms, the topic of the guide on clinical decision-making. Prognostic reasoning looks ahead: given this diagnosis, in this person, in this context, what is the most likely course and how wide is the range of possibilities.

What goes into a prognostic estimate

A prognostic estimate brings together three layers. The first is the natural history of the condition, what usually happens to those with that diagnosis. The second is the person: age, functional status, other conditions, support at home. The third is time: the prognosis changes as the condition evolves and must be redone.

Prognostic reasoning also has a product that diagnosis does not have: an interval. Saying "six months" is less honest than saying "probably between a few months and a year, with the greatest chance near six". The interval is part of the answer.

Prognosis is not only about survival

The daughter's question in the hallway is about length of life, but prognosis covers much more: the chance of regaining gait after a fracture, time until returning to work, risk of a new hospital stay, chance of an adverse effect. Much of the research on the accuracy of predictions, however, was done with survival in palliative care, and that is why the numbers in this guide come from that context.

Why prognosis disappeared from textbooks

The eclipse of prognosis

Nicholas Christakis analyzed how one disease, lobar pneumonia, was described in editions of a classic internal medicine textbook published between 1892 and 1988 (Christakis, 1997). The author called "the eclipse of prognosis" the process in which the topic lost explicit space in these texts (Christakis, 1997).

He points to two forces. The first was the emergence of effective treatment: the more therapy gains weight in managing a disease, the less prognosis appears, in an inverse relationship between the two clinical acts (Christakis, 1997). The second was a change in medical thinking itself, which moved from a view centered on the individual to a view centered on the diagnosis, in which prognosis seemed to come embedded in the name of the disease (Christakis, 1997).

What this leaves for the student

If the prognosis seems to come along with the diagnosis, nobody teaches it as separate reasoning. The student learns to arrive at the name of the disease and finds the prognosis on one line of the summary, as a fact to memorize. In practice, the patient wants to know what will happen to them, not to the average of those who have the same disease.

How often doctors get predictions right

Predictions that are too optimistic

In a prospective study across five outpatient palliative care services in Chicago, 343 doctors estimated the survival of 468 patients at the time of referral (Christakis and Lamont, 2000). Only 20% of the predictions fell within a margin of 33% of actual survival, 63% were too optimistic and 17% too pessimistic (Christakis and Lamont, 2000). On average, the doctors overestimated survival by a factor of 5.3 (Christakis and Lamont, 2000).

Two findings from the same study matter to those in training. The doctors in the highest quartile of experience were the most accurate (Christakis and Lamont, 2000). And accuracy fell as the relationship with the patient grew longer and the last contact more recent, a sign that emotional closeness pulls the estimate toward the optimistic side (Christakis and Lamont, 2000).

They get the size wrong, the order right

A systematic review gathered 1,563 pairs of prediction and actual survival in patients with terminal cancer (Glare et al., 2003). The predicted median was 42 days, and the actual one 29 days; the prediction was accurate within one week in 25% of the pairs and overestimated by at least four weeks in 27% (Glare et al., 2003).

Even so, the prediction and actual survival moved together: once transformed, the two had a correlation of 0.60 (Glare et al., 2003). The authors sum it up this way: the predictions discriminate, but they are miscalibrated (Glare et al., 2003). In other words, the doctor usually knows who is worse off, but gets wrong how much time each one has.

This distinction between discriminating and calibrating also appears in diagnosis. The confidence calibration guide shows how it applies to getting your own hypothesis right.

No group of specialists stood out

A more recent review included 42 studies of survival prediction in palliative care, with patients with and without cancer (White et al., 2016). The accuracy of estimates by category ranged from 23% to 78% across the studies (White et al., 2016). The four studies that asked for estimates in the form of a probability, such as "chance of being alive in one month", showed better performance, with a C statistic between 0.74 and 0.78 (White et al., 2016). And no professional group was consistently more accurate than the others (White et al., 2016).

The practical lesson is about format: thinking in terms of the probability of an outcome within a defined period tends to work better than trying to guess an exact number of days.

Simple prediction tools

The surprise question

The surprise question is one of the simplest tools: "would I be surprised if this patient died in the next twelve months?". It was designed to point out who may need palliative care. A systematic review with meta-analysis gathered 16 studies and 11,621 patients who answered this question (Downar et al., 2017).

For death between six and eighteen months, sensitivity was 67.0% and specificity 80.2% (Downar et al., 2017). The positive predictive value was 37.1% and the negative 93.1% (Downar et al., 2017). Performance was worse in diseases other than cancer, and the authors classified the question as a weak to modest predictor (Downar et al., 2017).

What these numbers teach

The surprise question is better for ruling out than for confirming: when the doctor answers "I would be surprised", the patient almost always is still alive within the period. When the doctor answers "I would not be surprised", the chance of death increases, but most patients are still alive at the end of the period. Reading sensitivity, specificity and predictive values is the same work as diagnosis, applied to a future outcome. The pretest probability guide explains these measures.

Scores and models

There are prognostic scores for many settings, from intensive care to oncology. They help because they provide a population anchor, similar to the base rate in diagnosis. But the review by Glare and colleagues showed that adding data such as functional status and symptoms to the clinical impression improved accuracy only slightly (Glare et al., 2003). The score is a starting point, and reasoning about the person is still necessary.

How to communicate an uncertain prognosis

Making uncertainty explicit

An article on teaching uncertainty notes that it usually appears implicitly, more suggested than stated, and argues for discussing it openly at all levels of training (Moulder et al., 2023). The authors recall that talking with the patient about diagnostic and prognostic uncertainty is a central part of shared decision-making (Moulder et al., 2023).

Uncertainty before diagnosis has its own guide: diagnostic uncertainty. In prognosis, it adds to another kind: even with the right diagnosis, each person's future varies.

Best case, worst case, most likely case

A review on prognosis in surgical oncology describes two techniques for situations of uncertainty: presenting the best case, the worst case and the most likely case, and proposing a treatment period with a defined deadline for reassessment (Fahy, 2019). Both replace a single number with a map of possibilities, which matches what studies show about the accuracy of predictions.

An example sentence

Compare two answers to the daughter's question. The first: "He has about six months." The second: "From what I see today, the most likely is that he has a few months. It could be less, if the next weeks are difficult, or more, if he responds well. We will reassess together at each visit." The second is longer, but it is truer to what the doctor knows.

How to train prognostic reasoning

Close the diagnosis and ask the third question

In every case you study, after reaching the diagnosis, write one line: "what do I expect to happen to this person in a month and in a year?". The habit is simple and brings prognosis back into the reasoning.

Write the prediction as a probability

Prefer "high chance of being at home in two weeks" to "will be discharged soon". Studies with estimates in probability performed better (White et al., 2016), and the format forces you to think about a timeframe.

Check the prediction later

During your rotation, note predictions about patients you follow: length of stay, response to treatment, need for readmission. Return to them later. It is the same principle as the error notebook: without comparing the prediction with what happened, no one finds out which way they err.

Distrust the optimism of those who know the patient

Accuracy fell the longer the relationship between doctor and patient (Christakis and Lamont, 2000). When you grow attached to a patient, ask a colleague who knows them less for an estimate. The metacognition guide brings other ways to watch your own judgment.

Discuss the prognosis with the preceptor

When presenting a case, include a prognosis sentence and ask the preceptor's opinion. The conversation reveals what experience adds, since the more experienced doctors were the most accurate in the study by Christakis and Lamont (Christakis and Lamont, 2000).

Prognosis and the case of the day in NEXO

In NEXO, each case leads to the diagnosis in up to six clues, and the ICD-10 proximity in five levels shows how far each guess was from the answer. It is a good starting point for the third question: after solving the case of the day, write in one line the course you expect for that person and discuss it with the class.

The cases are fictional with real clinical presentation, and the Archive, on the Modes screen, gathers previous cases for those who want to repeat the exercise. How the modes are unlocked is on the NEXO Economy page, and the what is NEXO guide explains the whole game.

If you coordinate a medical school, a student league or a residency and want to use the same case to discuss diagnosis and prognosis with the class, see NEXO Closer. Researchers interested in clinical reasoning can propose studies through NEXO Researcher. Doctors who review cases 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 prognostic reasoning?

It means estimating the likely course of a condition after diagnosis, with and without treatment, and expressing that estimate with the uncertainty it carries.

Do doctors usually get the prognosis right?

In palliative care, most survival predictions were too optimistic, and only one in five came close to actual survival (Christakis and Lamont, 2000).

If the predictions are wrong, what use are they?

They follow the order: patients the doctor thinks will live less tend to live less, even if the number of days is wrong (Glare et al., 2003).

What is the surprise question?

It means asking yourself whether you would be surprised if the patient died in the next twelve months. A meta-analysis rated it as a weak to modest predictor (Downar et al., 2017).

How do you give a prognosis without promising too much?

Use scenarios: best case, worst case and most likely case, with a date to reassess (Fahy, 2019).

How can NEXO fit into prognosis training?

The case of the day ends at the diagnosis, with up to six clues and the proximity in ICD-10. Use that ending as a starting point and write your prediction for the course of the case.

Prognostic reasoning checklist

1. Close the diagnosis and ask the third question: what will happen?
2. Separate the layers: natural history, person and time.
3. Write it as a probability with a defined time frame.
4. Give a range, not a single number.
5. Use the population as an anchor and adjust for the person.
6. Be wary of optimism when you know the patient well.
7. Communicate in scenarios and mark when to reassess.
8. Check afterwards what happened and note which way you were wrong.

Want to start with the first question? Play the case of the day on NEXO.

References

1. Christakis NA. The ellipsis of prognosis in modern medical thought. Soc Sci Med. 1997. https://pubmed.ncbi.nlm.nih.gov/9004366/
2. Christakis NA, Lamont EB. Extent and determinants of error in doctors' prognoses in terminally ill patients: prospective cohort study. BMJ. 2000. https://pubmed.ncbi.nlm.nih.gov/10678857/
3. Glare P, Virik K, Jones M, Hudson M, Eychmuller S, Simes J, Christakis N. A systematic review of physicians' survival predictions in terminally ill cancer patients. BMJ. 2003. https://pubmed.ncbi.nlm.nih.gov/12881260/
4. White N, Reid F, Harris A, Harries P, Stone P. A Systematic Review of Predictions of Survival in Palliative Care: How Accurate Are Clinicians and Who Are the Experts? PLoS One. 2016. https://pubmed.ncbi.nlm.nih.gov/27560380/
5. Downar J, Goldman R, Pinto R, Englesakis M, Adhikari NK. The "surprise question" for predicting death in seriously ill patients: a systematic review and meta-analysis. CMAJ. 2017. https://pubmed.ncbi.nlm.nih.gov/28385893/
6. Moulder G, Harris E, Santhosh L. Teaching the science of uncertainty. Diagnosis (Berl). 2023. https://pubmed.ncbi.nlm.nih.gov/36087299/
7. Fahy BN. Prognostication in oncology. J Surg Oncol. 2019. https://pubmed.ncbi.nlm.nih.gov/30883779/