Clinical article · Core, Acute care, Pathways and documents

The undifferentiated patient: from problem representation to a ranked differential

Written for
Hospital doctors, registrars and residents
Reading time
8 minutes
Published
Updated

Written by the Sana Health team for clinicians. It summarises published guidance, is not a guideline and has not been peer reviewed. Check current local protocols before acting.

Key points

  • Turn the history and findings into a one- or two-sentence problem representation before listing diagnoses.
  • Build the differential around the most discriminating finding, not the commonest symptom.
  • Test each candidate against its illness script and say explicitly what fits, what does not and what is missing.
  • Separate the diagnoses that must be excluded today from those that can be pursued in order.
  • A critical subset of diagnostic errors arises from cognitive errors such as premature closure [8]; a short, deliberate pause before committing to a diagnosis is a practical check.

Why the undifferentiated patient is hard

Many patients arrive without a diagnosis: breathless, confused, febrile or simply unwell, with findings that point in several directions. The diagnostic process runs from gathering information, through integrating and interpreting it, to generating hypotheses and confirming one, and each step is harder when the history is limited and the clinical, laboratory and imaging features are non-specific [1]. Diagnostic error is a recognised patient-safety problem across clinical settings, including the emergency department [2], and flawed clinical reasoning is associated with diagnostic errors and inappropriate investigations [3].

Experienced clinicians recognise many presentations quickly. Research on diagnostic reasoning describes two processes: a rapid, largely automatic retrieval of possible diagnoses based on experience, and a slower, analytical application of formal knowledge [4]. Neither is reliable on its own, and the high cognitive load of acute and intensive care makes both harder to use well [5]. A structured approach makes the analytical step explicit, so that it can be checked by the clinician, a colleague or a senior.

Start with a problem representation

A problem representation is a short summary that combines the facts of the case with the clinician's knowledge to describe the problem in medical terms [6]. It is usually one or two sentences. It names the context that changes the probabilities, the tempo of the illness and the clinical syndrome, and it uses semantic qualifiers: paired, abstract terms such as acute or chronic, focal or diffuse, intermittent or progressive [6, 7].

Compare two descriptions of the same patient. The first lists the facts: a 58-year-old man, vomiting for two days, pH 7.12, bicarbonate 9, glucose 9.8 mmol/L, ketones 5.4. The second is a problem representation: a middle-aged man taking an SGLT2 inhibitor, with an acute, severe ketoacidosis and a near-normal blood glucose. The second already points to a short differential and to what should be checked next.

  • Context: age, sex and the background that changes the probabilities, such as immunosuppression, pregnancy, recent travel or a new medicine.
  • Tempo: hyperacute, acute, subacute or chronic, and whether the course is stable, progressive or relapsing.
  • Syndrome: the pattern in medical terms, for example a high anion gap metabolic acidosis rather than abnormal bloods.
  • Key negatives: normal findings that make an important alternative unlikely.

Find the discriminating finding

Findings do not carry equal weight. Fatigue, nausea and a raised C-reactive protein have long lists of causes; a new pericardial rub, macroglossia or nephrotic-range proteinuria have short ones. Building the differential around the most specific finding keeps the list manageable and makes it more likely that the unifying diagnosis is on it.

Then ask whether a single process could explain the other findings too. Progressive exertional breathlessness, macroglossia, nephrotic-range proteinuria and a low-voltage ECG in the same patient should prompt consideration of amyloidosis rather than four separate problems. Medicines deserve particular attention at this stage: a drug effect can mimic disease in almost any organ system and is easily missed when the medication list is read as background.

Compare candidates with their illness scripts

An illness script is the clinician's stored picture of a disease: who gets it, how it evolves and what it looks like [6, 7]. Testing each candidate against its script, rather than against a list of symptoms, shows both what fits and what is missing. A feature the script predicts but the patient lacks is as informative as one that is present. For the man with ketoacidosis and a near-normal glucose, the comparison might look like this.

CandidateFitsAgainst, or still to checkWhat helps decide
Euglycaemic diabetic ketoacidosisTaking an SGLT2 inhibitor; severe ketoacidosis with a glucose lower than usually seen in DKAThe precipitant: intercurrent illness, reduced intake, a cut in insulin dose or recent surgeryRepeat ketones and venous blood gas; a search for the precipitant
Alcoholic ketoacidosisVomiting, ketosis, normal or low glucoseDepends on an alcohol history not yet takenAlcohol history, ethanol level, lactate
Starvation ketosisTwo days of vomiting and poor intakeUsually causes only a mild acidosis, not one this severeThe severity of the acidosis
Lactic acidosisA high anion gap acidosis can be mixedKetosis explains the picture only if the lactate is normalLactate; look for shock, sepsis, ischaemia or metformin
Toxic alcohol ingestionHigh anion gap metabolic acidosisDoes not explain the high ketones; ingestion not yet asked aboutOsmolar gap, history, advice from a poisons service

The table illustrates the method; it is not a management plan. Manage suspected diabetic ketoacidosis with the local protocol, including its guidance on SGLT2 inhibitors.

Decide what must be excluded today

A ranked differential is not the same as a plan. The most likely diagnosis is not always the most urgent, and a dangerous diagnosis of modest probability may need to be excluded before a likelier, benign one is pursued. Divide the list into what must be excluded or treated now and what can be investigated in order.

  1. Name the diagnoses that would cause harm if missed today, and the finding or test that would exclude each.
  2. Start treatment that is safe across the leading possibilities when waiting for certainty would cause harm.
  3. Order the remaining tests from quick, widely available and informative towards the invasive, and say what each result would change.
  4. Estimate how likely each leading diagnosis is before sending a test: a positive result when the pretest probability is very low is often a false positive, and a negative result does not exclude a diagnosis that was very likely.
  5. Record the plan and the reasoning, so the next clinician can see what has been excluded and what has not.

Guard against the common traps

A critical subset of diagnostic errors arises from cognitive errors: failures of perception, failed heuristics and biases [8]. A systematic review of studies in physicians found cognitive biases associated with diagnostic inaccuracies and management errors [9], and experimental studies show that context available during an encounter, such as how a case is framed, can change diagnostic accuracy [10]. These traps matter most in undifferentiated presentations.

  • Premature closure: stopping at the first diagnosis that explains most of the findings and leaving the rest unexplained.
  • Anchoring: holding on to an early impression after new information argues against it.
  • Diagnostic momentum: accepting a label from a referral letter, a previous admission or a handover without checking the evidence for it.
  • Discounting the result that does not fit: an unexplained finding is often the clue to the real diagnosis.
  • Chasing the rare: missing an atypical presentation of a common disease while looking for an unusual one.

Knowing the names of biases is not enough on its own. Diagnostic expertise rests heavily on knowledge, and how far teaching clinicians about biases reduces error is debated [4]. The practical safeguards work at the level of the case: a structured summary, an explicit check of what does not fit, and a second opinion when the picture stays unclear.

Pause before committing

Before a diagnosis is accepted and the plan handed over, a short, deliberate pause catches many of these problems. It takes a minute and fits into a ward round, a handover or a referral.

  1. Restate the problem representation and check that the leading diagnosis explains all of it.
  2. Ask what else this could be, and which dangerous alternative has not yet been excluded.
  3. List the findings that do not fit, and decide whether each needs an explanation.
  4. Say what result or change in the patient would make you abandon the leading diagnosis.
  5. Decide when the diagnosis will be reviewed, and what the patient and the next team should watch for.

When the case is still unclear, the usual responses apply: take the history again, re-examine the patient, look at the results and images themselves rather than only the reports of them, and ask a colleague or the relevant specialty.

How Sana Health helps

The Diagnostic Reasoning engine follows the same sequence. Describe the case, with results and their units, and it identifies the findings that narrow the differential most, flags time-critical problems first, ranks the differential with the features for and against each candidate, and sets out the investigations in order with what each would change. The report typically arrives in under a minute and ends with a numbered list of references read from PubMed and ClinicalTrials.gov when the question is asked. It supports clinical judgement; it does not replace it.

Questions

What is the difference between a problem representation and a case summary?

A summary lists the facts. A problem representation translates them into medical terms, adding the context, the tempo of the illness and the clinical syndrome, so that it points towards a differential. Breathless with swollen legs for three weeks is a description; subacute, progressive heart failure in a man with nephrotic-range proteinuria is a representation.

When is pattern recognition enough?

For a typical presentation of a common condition seen by a clinician with relevant experience, fast recognition is usually right and efficient. Slow down when the presentation is atypical, when findings remain unexplained, when the patient is not responding as expected, or when the cost of a missed diagnosis is high.

How should I document the reasoning?

Record the problem representation, the leading diagnoses with the main features for and against each, what has been excluded and how, the tests planned with the question each answers, and when the diagnosis will be reviewed. The next clinician can then pick up the reasoning rather than only the label.

References