Two single-item ratings from the clinician: how ill is the participant now, and how much have they changed since baseline. Short, public domain and used in almost every psychiatry trial, and increasingly beyond it.
Free sandbox · No credit card · 21 CFR Part 11 aligned
CGI-S: severity of illness
CGISCGI-I: global improvement since baseline
CGIIAt a glance
In trials
Disease-specific scales count symptoms. The CGI asks a trained clinician to integrate everything they know, including function and side effects, into one judgement. That makes it a useful check that changes on a symptom scale mean something clinically, and regulators often expect a global measure as a key secondary endpoint in psychiatric trials.
The CGI-S is collected at baseline and every assessment visit; the CGI-I from the first post-baseline visit onwards. Because the CGI-I compares with baseline, the rater needs access to the baseline condition, ideally by being the same person who rated it.
Patient Global Impression of Severity (PGI-S) and of Change (PGI-C) ask the participant the same kind of question. They are often used as anchors to interpret meaningful change on other patient-reported scales. See the MCID glossary entry.
Week 6 responders
38%
Assessed
84
Missing CGI-I
3
12% much or very much improved
Scales
| Score | CGI-S: severity | CGI-I: improvement |
|---|---|---|
| 1 | Normal, not at all ill | Very much improved |
| 2 | Borderline mentally ill | Much improved |
| 3 | Mildly ill | Minimally improved |
| 4 | Moderately ill | No change |
| 5 | Markedly ill | Minimally worse |
| 6 | Severely ill | Much worse |
| 7 | Among the most extremely ill | Very much worse |
Many protocols add a "not assessed" option coded separately from 0 so that missing ratings are explicit.
In Capture
The CGI is a short site form: dropdowns with the published labels, coded 1 to 7. The CGI-S is scheduled at every assessment visit and the CGI-I from the first post-baseline visit, and each rating is attributed to the person who entered it in the audit trail.
Study owner
Builds and configures the study
Principal investigator
Oversees the site, signs off the casebook
Sub-investigator
Enters and signs clinical data
Study coordinator
Screens subjects, enters visit data
Monitor (CRA)
Verifies data, raises queries
Data manager
Reviews, freezes and locks data
Participant
Completes their own questionnaires
Copy the template and schedule it at every assessment visit.
Beyond psychiatry
The CGI format has been adapted for many other conditions, from neurology to dermatology, usually by rewording the severity anchors for the disease. Adapted versions should define what each level means for that condition and train raters on the definitions, because "moderately ill" in autism and in Parkinson's disease describe very different pictures.
Whatever the condition, document the version and anchors in the protocol, keep the wording identical on every form, and treat the CGI as a clinician-reported outcome with the same rater discipline as longer scales such as the HAM-D.
Before go-live
Trained clinician, ideally the same one each visit.
CGI-I rated against the baseline visit.
CGI-I of 1 or 2, or another pre-specified rule.
Condition-specific definitions if not a psychiatric population.
PGI-S or PGI-C added if needed as an anchor.
Missing ratings explicit, not blank.
CGI-S at baseline and every assessment visit; CGI-I at every assessment visit after baseline.
It is usually a key secondary endpoint. In some conditions without a suitable disease-specific scale, CGI-based endpoints have been used as primary; discuss with regulators.
The Clinical Global Impression scale: clinician ratings of illness severity (CGI-S) and improvement since baseline (CGI-I), each from 1 to 7.
Yes. It was published by the US National Institute of Mental Health and is in the public domain.
A CGI-I of 1 (very much improved) or 2 (much improved) is a common responder definition.
No. CGI-I is rated relative to the participant's condition at baseline.
The CGI is rated by a clinician; the PGI asks the participant for their own global impression.
Yes, with condition-specific anchors that are defined and trained on.
Public-domain CGI, coded and attributed. Free sandbox.