CDIP study materials
CDIP study resources

Hospital Backlog Digitization With Mixed Paper and Handwriting

CDIP · Put your knowledge to work in context.

All case studies

Hospital Backlog Digitization With Mixed Paper and Handwriting

North Valley Regional Hospital is digitizing 1.2 million patient-record pages from 14 clinics before retiring its paper archive. The backlog includes 8.5 x 11 forms, 11 x 17 flow sheets, color-coded tabs, handwritten physician notes, and faded faxes. The scanning bureau must process 40,000 pages per day across three shifts with 22 operators, and HIPAA requires that every page be accounted for. The hospital wants full-text search for clinicians but insists the legal medical record image remain faithful to the original. Current samples show skew up to 12 degrees, staple holes, and OCR word accuracy of 91 percent on typed text and 62 percent on handwriting. The project manager must choose capture settings, quality sampling, and acceptance criteria without slowing throughput below the daily target.

1. Specify capture settings and image-enhancement rules for the mixed page types. How do you reconcile full-text search with the requirement that the legal medical record remain faithful?

Read the worked answer

Capture at 300 dpi for typed forms and 400 dpi for faded faxes and small print. Use 24-bit color for pages with color-coded tabs, annotations, or stamps; 8-bit grayscale for typed pages without color; and reserve bitonal for clean typed forms where storage or throughput demands it. Store the archival image as PDF/A-2 with lossless compression for bitonal and high-quality JPEG2000 for color, and run OCR on a derivative image so enhancement never alters the legal record. Deskew and despeckle only on the OCR derivative, and log any rotation or cropping applied to the archival image. Because handwriting OCR is 62 percent, do not rely on it for patient identity; use manual index fields for name, MRN, and date. Set OCR confidence at 80: below that, flag for human review. This preserves fidelity while delivering searchable text and keeps 40,000 pages per day achievable.

2. Design a quality-sampling plan that can run at 40,000 pages per day without becoming the bottleneck. What acceptance criteria and corrective actions would you set?

Read the worked answer

Use 100 percent page-count reconciliation against batch manifests, since HIPAA requires every page accounted for. For image quality, sample 5 percent of batches, about 2,000 pages at 40,000 per day, using a stratified plan: every operator, every scanner, and every document type each shift. Apply ANSI/ASQ Z1.4 general inspection level II with an AQL of 0.65 for major defects such as illegible text, missing pages, or wrong patient, and accept zero critical defects. Measure OCR accuracy on a 200-page typed sample per shift; if below 95 percent word accuracy, retrain the OCR engine or adjust resolution before continuing. Track skew, staple-hole false positives, and blank-page errors by operator and scanner. Corrective action: stop and recalibrate any scanner with three major defects in a sampled batch and rework the affected batch. Report daily defect rate and trend it on a control chart so throughput does not hide quality drift.

3. Faded faxes and handwritten notes are the highest-risk content. What capture and indexing decisions would you make for these, and what would you tell clinicians about search reliability?

Read the worked answer

For faded faxes, capture at 400 dpi grayscale, apply contrast normalization only to a search derivative, and retain the unenhanced archival image. If the fax is already degraded, do not promise OCR; instead index the document at the folder level and require manual keying of patient name, MRN, and document date. For handwritten notes, use manual indexing for the same fields and treat OCR output as a convenience, not the legal index. Set a rule that any page below OCR confidence 80, or any page classified as handwritten, enters a keying queue with a 24-hour turnaround. Tell clinicians that search covers typed content and indexed fields, not every handwritten word, and that the image is the authoritative record. This manages expectations, avoids false negatives in search, and keeps the legal record defensible.