Methods of Clinical Reasoning

This essay is by Jim Meadows, BScPT, MCPA, FCAMPT. It is the fourth in a series on clinical reasoning in physical therapy.

---

No single method of clinical reasoning works for every clinician, every patient, or every problem. The method that serves best depends on the clinician’s experience, the nature of the problem, and which signs and symptoms present first. Understanding the available methods — and their respective strengths and limitations — is what allows a clinician to choose wisely rather than default to habit.

The Main Methods of Clinical Reasoning

Five methods dominate clinical reasoning in physical therapy:

1. Algorithms — including flowcharts, treatment-based classification schemes, protocols, and triaging methods
2. Heuristics — experience-based rules of thumb (covered in the previous article in this series)
3. Pathognomonics — reasoning from a sign or symptom that is uniquely characteristic of a specific condition
4. Pattern recognition — the method of the expert, drawing on accumulated experience to reach rapid accurate conclusions
5. Hypothetico-deduction — the formal method of the non-expert, systematically generating and testing hypotheses

A sixth method, Script Focused Deduction (SFD), combines elements of pattern recognition and hypothetico-deduction into a practical framework accessible to novice clinicians. It is addressed in the first article in this series.

Pattern recognition, for example, can only be used when the presenting problem has been encountered in sufficient numbers for the clinician to recognize it without formal hypothesis testing — which is why it is mainly the domain of experienced practitioners. Pathognomonics is a method of opportunity: it is only available when a uniquely identifying sign or symptom is present.

Cognitive Bias in Clinical Reasoning

Alongside the choice of reasoning method, a second factor must be managed: cognitive error. The clinical reasoning literature documents approximately forty documented biases. The five most relevant to clinical diagnosis are:

1. Availability bias
2. Anchoring bias
3. Search satisfaction (premature closure)
4. Framing bias
5. Base rate neglect

These biases are unconscious and ingrained, which makes them difficult to manage by willpower alone. The most effective approach is to build bias correction into the reasoning process itself rather than relying on the clinician to recognize and override biases in the moment. This is addressed in detail in subsequent articles in this series.

Algorithms in Clinical Reasoning

Algorithms are step-by-step instructions for solving a problem. Originally developed in mathematics and later adopted in computer science, they have been applied to any number of non-mathematical tasks where an outcome can be reached through a defined sequence of discrete operations.

In clinical practice, algorithms are most commonly encountered as flowcharts, clinical prediction rules, and treatment-based classification schemes. Their appeal is straightforward: follow the steps correctly and you will arrive at a defined outcome. For someone with little experience, this is genuinely useful.

A Clinical Example: Post-Traumatic Headache

A flowchart algorithm for assessing post-traumatic headache for serious pathology (intracranial bleed, traumatic hydrocephalus, vertebrobasilar ischemia) might proceed as follows:

• Is the headache progressive? If yes → possible bleed → refer to physician.
• Are neurological symptoms present? If yes → refer to physician.
• Was onset sudden and severe? If yes → possible fracture → refer to physician.
• Is the headache deep, dull, and diffuse? If yes → possible bleed → refer to physician.
• Is dizziness present? If yes → are neurological signs present? If yes → possible neurological damage or VBI → refer to physician.
• Is the headache related to the dizziness and unrelated to the neck? If no → continue assessment.

This sequence works well within its scope. A novice clinician following it correctly will not miss a red flag presentation. The limitation is equally clear: if the patient’s problem falls outside the algorithm’s parameters, the algorithm cannot help. At that point, clinical reasoning must take over — and if the clinician has become dependent on the algorithm, they may not have the skills to do so.

The Strengths of Algorithmic Approaches

• They make relationships between signs, symptoms, and decisions visually clear
• They are relatively simple to follow within defined parameters
• When correctly designed, they eliminate bias and cognitive error from the decision
• They are effective teaching tools for understanding clinical processes and relationships

The Limitations of Algorithmic Approaches

• They can only handle relatively simple problems — as complexity increases, an algorithm either fails or becomes so elaborate it would be easier to reason through the problem directly
• When the algorithm fails, the clinician must be able to reason outside of it — but dependency on algorithms suppresses that ability
• Students who rely on algorithms may never develop the capacity to problem solve when the algorithm does not apply

Clinical Prediction Rules and Classification Schemes

Two widely used algorithmic approaches in physical therapy deserve specific attention: clinical prediction rules and treatment-based classification.

The clinical prediction rule for lumbar manipulation states that when certain criteria are met, the likelihood of manipulation providing meaningful pain relief reaches a specified level. Fewer criteria, lower probability. The rule is prognostic — it does not generate a diagnosis. The clinician must still reach the decision to manipulate through diagnostic reasoning, and then apply the rule to estimate likelihood of success.

This raises an immediate practical problem: if none of the criteria are met but the diagnosis still indicates manipulation, what does the clinician do? The rule does not answer that question. Without integration into a broader risk-benefit framework, its clinical utility is limited.

Treatment-based classification goes further — effectively replacing pathoanatomical diagnosis with a syndrome-based approach. The clinician collects subjective and objective findings, matches them to the best-fit category, and applies the predetermined treatment for that category.

The problems here are more fundamental. Most patients are more complex than a single category can accommodate — multiple conditions, interdependent pathologies, past treatment experience, and personality factors all affect assessment and management. Forcing a complex presentation into a category is more difficult than reasoning through it directly. More critically, if students use this method in straightforward cases where it functions adequately, they never develop the clinical reasoning skills needed for the cases where it does not.

When Algorithms Work Well: The Goldman Algorithm

The Goldman algorithm for triaging acute chest pain in the emergency department is a legitimate example of algorithms at their best. Goldman analyzed hundreds of acute chest pain presentations to identify the minimum set of findings most sensitive and specific for acute myocardial infarction: unstable angina, pulmonary edema, and systolic blood pressure below 100 mmHg.

He deliberately excluded predisposing factors — weight, family history, diabetes — that are essential for long-term cardiac management but are confounding noise for the immediate diagnostic question. Each criterion was weighted mathematically, and patients were sorted into one of three management categories. When tested at Chicago’s Cook County Hospital, the algorithm outperformed unaided physician judgment.

This works because it meets the conditions that make algorithmic approaches valid: it was built on rigorous research, applied to a relatively contained problem, used by trained physicians rather than taught as a substitute for reasoning, and functions as a triage tool — not a treatment protocol. It is the same category as the Ottawa Ankle Rules and the Canadian C-Spine Rules: decision aids for a specific narrow purpose, not replacements for clinical thinking.

The Goldman algorithm is also one of the key pieces of evidence supporting Script Focused Deduction: the finding that less information, carefully selected, can outperform more information gathered broadly, is central to the SFD framework.

Algorithms as Teaching Tools

Separate from their use as clinical protocols, flowcharts and algorithms have genuine value as learning aids. Drawing a flowchart of a clinical process — how signs and symptoms relate, what decisions they drive, what the consequences are — forces clarity of understanding. A teacher and students constructing a flowchart together surfaces gaps in reasoning. A student who can draw one from memory for a given condition has demonstrated understanding, not just recall.

Used this way, algorithms serve the development of clinical reasoning rather than replacing it.

Summary

Algorithms have legitimate strengths: they make clinical processes visual, they eliminate bias within their scope, and they are effective teaching tools. Their weaknesses are equally real: they cannot handle complex problems, they create dependency when used as primary clinical tools, and they tend to produce technicians rather than thinkers.

The clinician who understands algorithms well enough to use them purposefully — and to set them aside when the problem requires actual reasoning — is in a fundamentally different position from the clinician who depends on them. The goal of clinical education is to develop the latter into the former, not to provide protocols that make the transition unnecessary.

---

Also in This Series

• Script Focused Deduction: Mimicking the Expert
• Clinical Reasoning: Methods and Tools
• Heuristics and Axioms
• The Pathoanatomical Diagnosis
• Locking and Specificity in Spinal Manipulation