Most healthcare AI companies focus on imaging analysis or final diagnosis confirmation. We're taking a different approach: supporting clinicians at the very beginning of the diagnostic process—the moment a patient first presents with symptoms.

The Stages of Diagnosis

The diagnostic process unfolds in distinct stages:

1. Initial assessment – Patient presents with complaints; clinician rapidly considers possibilities
2. Differential generation – Clinician forms a mental list of potential diagnoses
3. Investigation ordering – Tests selected to narrow differentials
4. Results interpretation – Test results analyzed in context
5. Final diagnosis – Diagnosis confirmed; treatment initiated

Most AI tools target stages 4-5 (e.g., interpreting ECGs or chest X-rays). HeAIth targets stages 1-3—the earliest, most cognitively demanding phase.

Why the Start Matters Most

1. Diagnostic Errors Occur Early

Research from the Institute of Medicine and diagnostic safety studies consistently shows that most diagnostic errors stem from cognitive failures in the initial assessment phase—not from test misinterpretation or lab errors.

Common cognitive failures include:

• Anchoring bias – Fixating on an initial diagnosis and failing to consider alternatives
• Availability bias – Overweighting recent or memorable cases
• Premature closure – Stopping the diagnostic process too early
• Incomplete differential – Missing rare but serious diagnoses

2. Cognitive Overload is Highest at the Start

When a patient says "I have chest pain and shortness of breath," a clinician's brain must rapidly consider:

• Cardiac causes (MI, angina, pericarditis, aortic dissection)
• Pulmonary causes (PE, pneumonia, pneumothorax, pleurisy)
• Gastrointestinal causes (GERD, esophageal spasm)
• Musculoskeletal causes (costochondritis)
• Anxiety/panic disorders

This happens in seconds, often while managing interruptions, time pressure, and incomplete patient history.

3. The First Investigations Determine the Path

The initial tests ordered often determine the entire diagnostic trajectory. Order a troponin and ECG, and you're on a cardiac pathway. Order a d-dimer and chest X-ray, and you're considering PE. Order nothing, and you might miss something critical.

Getting the initial workup right matters enormously—both for patient safety and healthcare efficiency.

Where AI Can (and Can't) Help

What AI is Good At:

• Comprehensive pattern matching – Considering all possible diagnoses from symptom combinations
• Mitigating bias – Not swayed by recent cases or gut feelings
• Evidence-based recommendations – Suggesting investigations based on clinical guidelines
• Flagging red flags – Identifying high-risk symptom combinations requiring urgent action

What AI is Bad At:

• Clinical context – Understanding the patient's social situation, health literacy, or ability to follow up
• Physical examination – Reading body language, facial expressions, or subtle physical findings
• Shared decision-making – Negotiating diagnostic plans with patients based on values and preferences
• Judgment calls – Knowing when to "watch and wait" vs. "investigate aggressively"

This is why HeAIth supports, rather than replaces, clinical judgment. We provide comprehensive differential suggestions and evidence-based investigation recommendations. The clinician makes the final call.

Clinical Scenarios Where Early Support Matters

Scenario 1: The Busy GP Surgery

Context: 10-minute appointment, patient presents with vague fatigue and weight loss.

Without HeAIth: GP might anchor on depression or thyroid issues, order basic bloods, miss red flags for malignancy.

With HeAIth: AI flags weight loss + fatigue as requiring broader workup. Suggests comprehensive blood panel including inflammatory markers, highlights red flags requiring urgent investigation.

Scenario 2: The A&E Triage

Context: Patient presents with abdominal pain. Multiple patients waiting, high cognitive load.

Without HeAIth: Clinician focuses on appendicitis, orders abdominal ultrasound, misses atypical presentation of inferior MI.

With HeAIth: AI suggests abdominal pain + nausea can be cardiac. Recommends troponin and ECG alongside abdominal workup.

Scenario 3: The Rare Presentation

Context: Young patient with headache and neck stiffness, recently had "flu."

Without HeAIth: Clinician attributes symptoms to viral illness, reassures patient, sends home.

With HeAIth: AI flags headache + neck stiffness + recent illness as possible meningitis. Recommends urgent evaluation and LP consideration.

What HeAIth is NOT

To be clear about our scope:

• Not a diagnostic tool – We don't provide final diagnoses; we assist with differential generation
• Not a replacement for clinical judgment – The clinician always makes the final decision
• Not imaging AI – We don't interpret scans or ECGs; we help decide which tests to order
• Not a symptom checker – Patients shouldn't use this; it's designed for trained clinicians

The Evidence Base

Our approach is grounded in diagnostic safety research:

• Graber et al. (2005): "Cognitive factors account for 74% of diagnostic errors"
• Singh et al. (2013): "Missed and delayed diagnoses are the largest category of malpractice claims in primary care"
• National Academies (2015): "Improving Diagnosis in Health Care" – Emphasizes the need for diagnostic support tools
• NICE guidance on clinical decision support systems (2019)

Why This Matters for the NHS

The NHS faces unique pressures:

• GP appointment times – 10 minutes to assess, diagnose, and treat
• A&E overcrowding – High cognitive load, interruptions, handovers
• Workforce shortages – Junior doctors covering multiple specialties
• Diagnostic delays – Cancer referral targets often missed due to late recognition

AI support at the start of diagnosis can help clinicians:

• Reduce diagnostic errors and missed diagnoses
• Order appropriate initial investigations
• Identify red flags requiring urgent action
• Reduce cognitive load and burnout

Join the Conversation

We're building HeAIth with frontline clinicians, not for them. If you have thoughts on how AI can best support early diagnostic decision-making, we'd love to hear from you.