A Drug Requiring Prior Authorization Is Prescribed
A prescribed drug triggers a required insurer prior authorization check.
- ~35%: Drugs requiring PA (of specialty prescriptions)
- 2–5 days: Avg PA processing time (per insurer request)
- 13 hrs/wk: Physician hours on PA (AMA survey average)
- 93%: Care delayed by PA (of surveyed physicians report)
Why prior authorization exists
Insurers require proof a drug is medically necessary.
The prediction opportunity
Predicting outcomes early avoids costly, slow denials.
What the AI observes
Drug, diagnosis, dosage, and payer feed the model.
Prediction happens before submission, not after denial.
AI Evaluates Diagnosis-Code Match And Documentation
AI scans diagnosis codes and clinical notes for completeness.
- ICD-10: Diagnosis-code check (vs payer policy criteria)
- 40+: Documentation fields scanned (chart and note elements)
- ~24%: Missing-doc denial share (of initial PA denials)
- <10 sec: Assessment time (per request scanned)
Diagnosis-code matching
The AI checks whether ICD-10 codes support the drug.
Documentation completeness scoring
Missing notes, labs, or history lower the score.
Why documentation drives denials
Incomplete charts are a leading cause of denial.
A 20% documentation gap often predicts denial.
AI Compares Against Similar Past Decisions
The model compares this request against thousands of past cases.
- 100k+: Historical cases referenced (prior approval/denial records)
- 12: Similarity features compared (drug, diagnosis, payer, dosage)
- ~88%: Pattern-match accuracy (against payer outcomes)
- 500+: Payer policy variants tracked (insurer-specific rule sets)
Learning from precedent
The model studies thousands of similar past requests.
Clustering by similarity
Cases with matching features cluster into outcome groups.
Payer-specific patterns
Approval patterns vary sharply by insurer and plan.
Nearby historical cases strongly shape the prediction.
Approval Probability Is Estimated Before Submission
A calibrated probability estimates likely insurer approval or denial.
- 0–100%: Predicted approval probability (calibrated confidence score)
- 0.86: Model AUC (validation) (approval prediction benchmark)
- 3: Key predictive factors (diagnosis match, docs, history)
- <1 sec: Calculation latency (before submission)
Combining the signals
Match strength, completeness, and history blend into one score.
Calibrated, not just confident
The score reflects real-world approval frequency, not guesswork.
A decision-support number
Clinicians see a probability, not an automatic decision.
Higher completeness reliably lifts predicted approval odds.
Prediction Shown With Improvement Suggestions
Clinicians see the odds plus concrete ways to improve them.
- ~30%: Denials potentially avoided (with pre-submission fixes)
- 1–3: Suggested doc additions (per flagged request)
- 3: Outcome categories shown (likely / uncertain / likely denied)
- ~2 days: Faster resubmission cycle (saved on average)
Reading the prediction
Likely Approved, Uncertain, or Likely Denied is displayed.
Actionable suggestions
The AI flags exactly which documentation to add.
Closing the loop
Better documentation before submission raises real approval rates.
Fixing flagged gaps can flip a denial into approval.