Christopher LyversAI Quality & Data Analyst
← All work samples

AI Evaluation / Evidence Reconciliation / QA Work Sample

Claims Evidence Audit

Denied-date analysis & record exclusion

Audited 14 indexed records across a 19-day denied range, excluded contaminated evidence, separated defensible dates from weak ones, and removed unsupported diagnoses.

Scope

14indexed records across audit, claim and care documentation
19days in the disputed 01/12–01/30 range
3records requiring exclusion or reduced evidentiary weight
3risk tiers for the final findings

01

Overview

This simulated skilled-nursing claims assignment asked whether an appeal narrative was actually supported by the underlying record. I reconciled identity and dates, compared contemporaneous evidence with a flawed draft, separated stronger days from weaker ones, and produced narrow language that did not outrun the source documentation.

02

Challenge

The package deliberately mixed valid evidence with a wrong resident's hospital record, a same-resident document outside the denied period, a late unsigned addendum and an internal draft containing overstated diagnoses. The date range also changed character: early notes showed measurable skilled needs, while later records became increasingly routine.

03

Source material

  • Audit request and claim summary
  • Hospital discharge and skilled-nursing admission records
  • Physician/practitioner and nursing notes
  • Therapy evaluations and daily logs
  • MDS/care-plan excerpt and medication/lab/order records
  • Internal draft appeal plus identity, date-range and signature traps

04

My work

  • Matched resident and claim identifiers before using any document.
  • Excluded the wrong-resident record and prevented its diagnoses from contaminating the appeal analysis.
  • Separated contemporaneous denied-range documentation from an outside-date consult and a late unsigned therapy addendum.
  • Reviewed the disputed period day by day for skilled versus routine or custodial content.
  • Compared the internal draft's claims with the actual diagnoses, signatures, dates and service records.
  • Produced supported appeal language, a missing-document list and high/medium/lower risk rankings.

05

Analytical approach

01

Validate resident and claimCross-check name, identifiers, stay dates and denied dates before considering substance.

02

Assign evidence weightExclude wrong identity, isolate outside-range material and downgrade late or unsigned records.

03

Analyze by dateAvoid a single all-or-nothing conclusion for a period whose evidence strength changes over time.

04

Audit the draftTrace every diagnosis and service claim back to contemporaneous records.

05

Write narrowlyRecommend only language that survives identity, timing, signature and source-support checks.

06

Key findings

Specific examples from the completed work.

Record contamination prevented

A hospital record belonged to Ellen M. Barnes, not Eleanor J. Barnes, with different identifiers and dates. It was excluded. A 02/06 wound consult was after discharge and outside the denied range; a 02/10 therapy addendum was late and unsigned.

Date-specific support

The strongest support was 01/12–01/16, when records documented anticoagulation adjustment, cardiopulmonary monitoring during oxygen weaning, wound care, and skilled gait, endurance, transfer, ADL and safety work with measurable deficits.

Later period was weaker

The evidence from 01/17–01/30—especially 01/19 onward—shifted toward routine medications, ADL supervision, stable findings, group exercise, home-program review and discharge preparation, with gaps in daily therapy documentation.

Unsupported draft claims removed

The records did not support claims of sepsis, acute stroke, skilled speech therapy for dysphagia, IV antibiotics through discharge, a stage 3 ulcer during the stay, wound-vac care, oxygen dependence through discharge or hands-on help with all ADLs through 01/30.

Documentation risk made explicit

Missing or late signatures, no signed recertification after 01/15, and delayed discharge documentation were separated from the underlying evidence of skilled need.

07

Deliverable

A concise appeal-support review identifying defensible and weak dates, excluded evidence, unsupported statements, timing/signature/coding concerns, missing records to obtain, narrowly supported language, and a three-level risk ranking.

08

Skills demonstrated

  • Claims-document QA
  • Identity reconciliation
  • Date-range analysis
  • Evidence inclusion/exclusion
  • Unsupported-claim detection
  • Risk classification
  • Draft-to-source audit
  • Research synthesis

Why it matters

The assignment maps closely to data-quality and research work: establish which records are admissible, weight evidence by recency and reliability, identify narrative overreach, and communicate a conclusion that is useful precisely because it is bounded.