Common omissions in chart summaries, and a checklist
By Ahmed Jemaa, Co-Founder and CEO of Medrecords AI. Updated .
Short version: the items most often missing from discharge summaries are patient and family counseling (90% to 92%), test results pending at discharge (65%), diagnostic test results (33% to 63%), follow-up plans (2% to 43%) and discharge medications (2% to 40%), per a 2007 JAMA systematic review. AI drafts add a new failure: statements the chart doesn't support. Use the checklist below on any summary before you send or sign it.
Limits: Vendor facts come from each vendor's own website on 2026-11-12 and were not tested by us. Rules are summarized from the cited text; your hospital's bylaws, payer contracts and state law can ask for more.
What the research found missing
The best data are on discharge summaries. Kripalani and colleagues pooled studies of communication between hospital physicians and primary care physicians (JAMA, 2007). Their ranges are spreads across the pooled studies.
Figure 1. Where discharge summaries come up short. Data: Kripalani et al., JAMA 2007;297(8):831-41, abstract read on PubMed 2026-11-12. Chart drawn by Medical Chart Summary.
Finding
Number
Study
Patients discharged with test results still pending
41% (1,095 of 2,644 patients, 2,033 results)
Roy et al., Ann Intern Med 2005, 2 academic hospitals
Of those results, potentially actionable
9.4%
Roy et al., 2005
Physicians unaware of the surveyed actionable results
61.6%
Roy et al., 2005
Admitted patients (on 4 or more drugs) with at least 1 unintended medication discrepancy
53.6% of 151
Cornish et al., Arch Intern Med 2005
Most common discrepancy
A regular medication left out (46.4%)
Cornish et al., 2005
Discrepancies that could cause moderate to severe discomfort or deterioration
38.6%
Cornish et al., 2005
Direct physician-to-physician communication at discharge
3% to 20%
Kripalani et al., JAMA 2007
All abstracts read on PubMed on 2026-11-12. These studies are 2 decades old. Newer EHR data may differ, but we found no larger study that replaces them.
The new omission: confabulation in AI drafts
AI drafts change the failure pattern. In a 2025 study at NYU Langone, hospitalists compared AI-drafted and physician-written hospital courses after residents edited both under time limits. The AI drafts needed less editing (31.5% of text changed versus 44.8%) and were rated more complete, but held more confabulations (Small et al., JAMA Netw Open 2025). A summary can now be complete and wrong at the same time. So the checklist has 2 halves: what is missing, and what is unsupported.
Omissions by format: the checklist
Pick the format, tick each item the summary covers, and the tool lists what is left. It runs in your browser; nothing is sent anywhere. Items marked "high risk" are the ones the research above found missing most often.
Not covered yet: all items.
Why the same items keep slipping
Pending results belong to no one by default. The ordering team leaves; the receiving team doesn't know the test exists. The fix is a name next to every pending test.
Medication lists show state, not change. A list of current doses hides every change made during the stay. Verbs fix it.
Incidental findings aren't diagnoses. They sit in an imaging report, outside the problem list, and nothing pulls them forward. Put them in their own field.
Counseling happens out loud. If no one writes down what the patient was told, the next clinician can't build on it.
The worked examples show each of these on 1 synthetic chart, and the templates have a field for each.