Primary Care AI Standard of Care: Clinical Decision Support, Diagnostics, and Liability

AI Enters the Primary Care Practice

#

Primary care represents perhaps the most consequential frontier for artificial intelligence in medicine. As the first point of contact for most patients, primary care physicians face the challenge of distinguishing serious conditions from benign presentations across every organ system, managing complex chronic diseases, and coordinating care across specialists, all while seeing 20-30+ patients per day. AI promises to enhance diagnostic accuracy, improve chronic disease management, and catch the “needle in a haystack” diagnoses that might otherwise be missed. But with this promise comes significant liability questions: When an AI clinical decision support system fails to suggest a diagnosis that a prudent physician should have considered, who is responsible?

This guide examines the standard of care for AI use in primary care, the expanding landscape of FDA-cleared diagnostic tools, and the emerging liability framework for AI-assisted ambulatory care.

FDA-Cleared Primary Care AI Devices

#

Clinical Decision Support Systems

#

The largest category of primary care AI involves diagnostic assistance:

Major FDA-Cleared Diagnostic Support Systems (2024-2025):

Point-of-Care Diagnostics

#

AI-enhanced testing for primary care settings:

Applications:

• Diabetic retinopathy screening
• Dermatology lesion analysis
• ECG interpretation
• Urinalysis interpretation
• Point-of-care ultrasound

Major Devices:

Chronic Disease Management

#

AI for ongoing care of chronic conditions:

Applications:

• Diabetes management and insulin dosing
• Hypertension monitoring and treatment optimization
• COPD exacerbation prediction
• Heart failure remote monitoring
• Medication adherence prediction

Recent FDA Clearances:

• Apple Hypertension Notification Feature (September 2025)
• Various glucose monitoring systems with AI prediction
• Remote patient monitoring platforms

Preventive Care and Screening

#

AI supporting population health:

Applications:

• Cancer screening risk stratification
• Cardiovascular risk prediction
• Social determinants of health identification
• Preventive care gap identification
• Vaccine recommendation systems

The Liability Framework

#

The Diagnostic Error Challenge

#

Primary care faces unique diagnostic pressures:

The Problem:

• Undifferentiated presentations (fatigue, pain, malaise)
• Low disease prevalence makes serious conditions rare but important
• Limited time for each encounter
• Broad scope across all organ systems
• Follow-up often depends on patient return

The Central Question:

“If AI could have suggested a diagnosis that the physician didn’t consider, and the patient was harmed by the delay, does failure to use available AI tools constitute negligence? Conversely, if a physician relies on AI that fails to flag a serious condition, is that reliance reasonable?”

The Learned Intermediary Doctrine

#

Traditional Framework:

• Physicians are “learned intermediaries” between products and patients
• AI clinical decision support is a tool, not a substitute for judgment
• Manufacturer’s duty is to adequately warn the physician
• Physician’s duty is to apply clinical judgment

AI Complications:

• AI may “know” more than any individual physician
• Should physicians be expected to use available AI?
• When AI and physician disagree, whose judgment prevails?
• How detailed must AI warnings about limitations be?

Liability Allocation

#

Primary Care Physician Responsibility:

• AI is advisory, not determinative
• Must maintain independent diagnostic capability
• Cannot delegate pattern recognition entirely to AI
• Document reasoning for agreeing/disagreeing with AI
• Understand AI limitations in your patient population
• Ensure appropriate follow-up regardless of AI output

Device Manufacturer Responsibility:

• Clear labeling of intended use and limitations
• Training on appropriate clinical scenarios
• Transparency about false negative/positive rates
• Post-market surveillance for missed diagnoses
• Timely safety communications

Health System Responsibility:

• Validation before deployment in primary care
• Training for clinical staff
• Integration into EHR workflow
• Quality monitoring and outcome tracking
• Ensuring AI doesn’t introduce new workflow risks

Clinical Applications and Risk Areas

#

Diagnostic Decision Support

#

The Value Proposition:

• Average primary care visit: 12 minutes face-to-face time
• Physicians can’t consider every possible diagnosis
• AI can suggest diagnoses physician might not have considered
• Particularly valuable for rare conditions

AI Role: Systems like Isabel DDx and DXplain analyze patient symptoms, history, and test results to suggest differential diagnoses the physician might consider.

Liability Concerns:

AI Doesn’t Suggest Correct Diagnosis:

• Patient presents with vague symptoms
• AI generates differential that doesn’t include ultimate diagnosis
• Physician, informed by AI output, doesn’t consider the condition
• Delayed diagnosis with patient harm

AI Suggests Diagnosis But Physician Doesn’t Act:

• AI includes serious condition in differential
• Physician dismisses as unlikely
• No testing or follow-up arranged
• Patient returns with advanced disease

Case Pattern: Missed Cancer 55-year-old presents with fatigue and weight loss. AI clinical decision support generates differential including depression, thyroid disease, and (further down the list) malignancy. Physician focuses on depression and thyroid, orders TSH. Patient returns 4 months later with advanced pancreatic cancer. Question: Was failure to pursue the AI-suggested malignancy workup negligent?

Diabetic Retinopathy Screening

#

The Stakes:

• Diabetic retinopathy is leading cause of blindness in working-age adults
• Annual screening recommended but only ~60% of diabetics screened
• Early detection prevents 95% of vision loss
• Point-of-care AI screening could close the gap

AI Solution: IDx-DR was the first FDA-authorized autonomous AI diagnostic, it can diagnose diabetic retinopathy without physician interpretation. Primary care can screen patients during routine visits.

Liability Considerations:

• If AI misses retinopathy that progresses to vision loss
• If AI creates false positive leading to unnecessary referral
• Question of whether AI screening is now standard of care
• Responsibility when AI detects but patient doesn’t follow up

Autonomous AI Implications: IDx-DR’s autonomous designation changes liability analysis:

• Device provides diagnosis, not just information
• Manufacturer may bear more direct liability
• But physician must ensure appropriate use and follow-up
• Patient selection (image quality, exclusions) still matters

Skin Cancer Detection

#

AI Applications:

• DermaSensor: FDA-cleared device for skin cancer detection
• SkinVision: Smartphone app for lesion assessment
• AI dermoscopy analysis
• Mole mapping with AI tracking

Liability Issues:

• False negative: Patient reassured, cancer progresses
• False positive: Unnecessary biopsy with complications
• Scope of practice: Primary care using dermatology AI
• Consumer apps vs. medical devices

The Referral Question: Should primary care physicians use AI to determine which lesions need dermatology referral? Or does AI create obligation to refer anything concerning?

Cardiovascular Risk Assessment

#

AI Applications:

• Enhanced cardiovascular risk calculators
• ECG-based AI for subclinical disease (AFib, low EF)
• AI analysis of lipid panels for familial hypercholesterolemia
• Social determinants integration for risk prediction

Liability Considerations:

• AI identifies high risk but physician doesn’t intensify treatment
• AI misses high-risk patient due to atypical presentation
• Overtreatment based on AI risk scores
• Patient autonomy when AI predicts high risk

Mental Health Screening

#

Emerging AI:

• Depression screening with natural language processing
• Suicide risk prediction
• Anxiety disorder identification
• Substance use disorder risk assessment

Unique Liability Issues:

• Privacy concerns with mental health AI
• Duty to act on AI-identified suicide risk
• False positives creating stigma or unnecessary intervention
• Integration with primary care workflow

Professional Society Guidelines

#

AAFP Position on AI (2024)

#

The American Academy of Family Physicians has addressed AI:

Key Principles:

Clinical Decision Support:

• AI should enhance, not replace, clinical reasoning
• Physicians must understand AI capabilities and limitations
• AI output is one input to clinical judgment
• Documentation should reflect AI use and physician assessment

Training and Competency:

• Medical education must include AI literacy
• Continuing education on AI tools essential
• Understanding of AI limitations critical
• Competency in underlying clinical skills must be maintained

Implementation:

• Validation in diverse patient populations
• Integration without workflow disruption
• Quality monitoring for AI-assisted care
• Equity considerations (does AI perform equally across demographics?)

AMA Guidance on AI in Practice

#

Key Recommendations:

• AI should be fair, safe, effective, and transparent
• Physician oversight of AI is essential
• AI should address health inequities, not worsen them
• Data privacy must be protected
• Liability framework should be clarified

Joint Commission Standards

#

Relevant Standards:

• Clinical decision support systems require validation
• Staff training on AI tools required
• Quality monitoring must include AI performance
• Adverse events related to AI must be reported

Standard of Care for Primary Care AI

#

What Reasonable Use Looks Like

#

Diagnostic Support:

• Use AI as one input to differential diagnosis
• Consider AI suggestions as prompts for clinical reasoning
• Don’t use AI as a substitute for thorough history and exam
• Document when AI suggestions are adopted or rejected
• Ensure appropriate follow-up regardless of AI output

Point-of-Care Testing:

• Use AI diagnostics according to FDA-cleared indications
• Understand sensitivity/specificity in your population
• Ensure appropriate referral for positive results
• Don’t over-rely on negative results when clinical suspicion is high

Chronic Disease Management:

• AI can help identify patients needing intervention
• Treatment decisions remain with physician
• Patient preferences must be incorporated
• Monitor outcomes of AI-guided care

What Falls Below Standard

#

Diagnostic Failures:

• Using AI as substitute for clinical assessment
• Ignoring AI-suggested serious diagnoses without reasoning
• Not understanding AI limitations in atypical presentations
• Failing to ensure follow-up when diagnosis uncertain
• Over-relying on AI “normal” results

Implementation Failures:

• Deploying AI without validation in your patient population
• No training for clinical staff
• Using AI outside cleared indications
• No quality monitoring

Documentation Failures:

• Not recording AI use in clinical decisions
• No documentation of reasoning when AI suggestions rejected
• Missing follow-up plans when AI inconclusive

Malpractice Considerations

#

The Diagnostic Failure Pattern

#

Primary care malpractice often involves delayed or missed diagnosis:

Traditional Elements:

1. Patient presents with symptoms
2. Physician fails to diagnose condition
3. Delay leads to disease progression
4. Patient suffers harm from delayed treatment

AI Adds New Questions:

• Was AI clinical decision support available?
• Did AI suggest the correct diagnosis?
• Did physician consider and document AI output?
• Would AI have caught what physician missed?

The “Should Have Used AI” Argument

#

Emerging Plaintiff Theory:

• AI diagnostic tools were available
• AI would likely have suggested correct diagnosis
• Failure to use available tools was negligent
• Patient harmed by the omission

Defense Responses:

• AI use not yet standard of care
• AI has its own error rates
• Clinical judgment appropriately applied
• AI not validated for this presentation

Defense Strategies

#

For Primary Care Physicians:

• Document clinical reasoning independent of AI
• When using AI, document its suggestions and your response
• Note AI limitations relevant to patient
• Ensure appropriate follow-up
• Show competency in underlying clinical skills

For Health Systems:

• Validation documentation
• Training records
• Quality monitoring showing AI performance
• Protocol development and compliance
• Adverse event tracking and response

For Manufacturers:

• FDA clearance as evidence of safety
• Proper labeling and warnings
• Training program adequacy
• Known limitations disclosure
• Post-market surveillance compliance

Implementation Considerations

#

EHR Integration

#

Critical Factors:

• AI must integrate seamlessly into workflow
• Alert fatigue risk if too many AI notifications
• Must not slow down already-pressed encounter time
• Documentation must be efficient

Liability Implications:

• Poor integration may cause AI to be ignored
• Workflow disruption may introduce new errors
• Alert fatigue may cause missed serious warnings
• Documentation requirements must be realistic

The Time Pressure Reality

#

Primary Care Context:

• 12 minutes average face-to-face time
• 20-30 patients per day
• Extensive documentation requirements
• Administrative burden already high

AI Promise vs. Reality:

• AI should save time, not add burden
• Complex AI outputs may slow decisions
• Learning curve for new AI tools
• Risk of shortcuts if AI adds complexity

Equity Considerations

#

AI Bias Risks:

• Training data may not represent your patient population
• Performance may vary across demographic groups
• Social determinants may not be adequately captured
• Language and cultural factors may affect accuracy

Validation Requirements:

• Test AI performance in your specific population
• Monitor for disparities in AI recommendations
• Ensure equity in AI-assisted care
• Report concerns about AI bias

Frequently Asked Questions

#

Related Resources

#

AI Liability Framework

#

• AI Misdiagnosis Case Tracker, Diagnostic failure documentation
• AI Product Liability, Strict liability for AI systems
• Radiology AI Standard of Care, Diagnostic imaging AI

Healthcare AI

#

• Healthcare AI Standard of Care, Overview of medical AI standards
• AI Medical Device Adverse Events, FDA MAUDE analysis
• Cardiology AI Standard of Care, Cardiovascular AI liability

Emerging Litigation

#

• AI Litigation Landscape 2025, Overview of AI lawsuits