CGI: Clinical Global Impressions Scale

What this assessment is

The Clinical Global Impressions Scale, or CGI, is a brief clinician-rated framework for summarizing illness severity, change, and treatment effect.

The original CGI has three separate components:

• CGI-Severity, often written CGI-S
• CGI-Improvement, often written CGI-I
• CGI Efficacy Index

CGI-S and CGI-I each use a rating from 1 to 7. A code of 0 means not assessed.

The Efficacy Index uses a matrix that considers therapeutic effect and adverse effects together.

These components do not add to one total.

Keep them separate.

Who it was designed for

The CGI appeared in the 1976 ECDEU Assessment Manual for Psychopharmacology. It was designed for clinicians evaluating people in mental health treatment and research.

The framework is not tied to one diagnosis. Studies have used it in depression, anxiety, psychosis, bipolar disorder, and other conditions.

That breadth is useful, but it creates a requirement: define what illness the rating concerns.

Name the target.

A CGI score is hard to interpret if the record does not name the target condition, baseline, timeframe, and evidence available to the rater.

The rater needs enough clinical knowledge to judge the whole picture. CGI is not a patient self-report scale.

Clinical judgment remains central.

How it is administered

A clinician reviews the relevant symptoms, functioning, behavior, treatment effects, and other clinical information.

For CGI-S, the clinician rates current overall illness severity. For CGI-I, the clinician compares the current state with a stated baseline, usually the start of treatment or study entry.

Baseline matters.

The Efficacy Index requires separate judgments about therapeutic benefit and adverse effects.

Details matter.

The assessment can be quick after a full clinical encounter. It should not replace that encounter.

Brevity is not certainty.

Before rating, define:

• the target condition
• the assessment date and timeframe
• the comparison baseline for CGI-I
• the information sources used
• whether another clinician made earlier ratings

Define these points first.

Rater training and calibration improve consistency. Two clinicians can otherwise use the same number for different clinical states.

What it measures

CGI-S summarizes current illness severity in one global judgment.

CGI-I summarizes change from baseline. It asks whether the person's condition is better, unchanged, or worse overall.

The Efficacy Index combines the degree of therapeutic benefit with the burden of adverse effects. It is not simply another severity score.

These ratings compress a large amount of information into a small number. They do not show which symptom, behavior, or area of function changed.

Detail is lost.

The GAF is a historical 1-to-100 rating that combines symptoms and functioning. CGI keeps current severity and change separate, but it offers less detail than a disorder-specific measure.

Use CGI with a condition-specific symptom scale and a functional measure when detail matters.

How scoring works

CGI-S ranges from 1 to 7. Higher numbers indicate greater clinician-rated illness severity.

CGI-I also ranges from 1 to 7. Lower numbers indicate improvement, 4 means no change, and higher numbers indicate worsening.

A score of 0 on either component means not assessed. It is not a clinical result and should not enter averages as the lowest possible score.

The Efficacy Index uses categories for therapeutic effect and adverse effects. The rater selects their intersection in the official matrix.

Do not add CGI-S, CGI-I, and Efficacy Index values. Do not create a total or percentage.

Do not combine them.

Keep each component labeled. A bare value of 2 is ambiguous because it has a different meaning on CGI-S and CGI-I.

Labels matter.

No general missing-data or prorating rule is needed. If the clinician cannot make a supported judgment, record not assessed.

Use zero honestly.

How to interpret a CGI score

Interpret CGI-S as a summary of current severity for the named condition and timeframe.

Interpret CGI-I as change from the named baseline. A value cannot show change unless the baseline is clear.

Record it.

A shift in CGI-S is not identical to a CGI-I rating. CGI-S compares the current state with the rater's concept of illness severity. CGI-I compares the current state with one person's earlier state.

There is no universal minimum important change that applies across all diagnoses, settings, and raters.

Do not invent one.

Because the rating compresses symptoms, function, adverse effects, and clinical judgment, a one-point shift needs an explanation before it guides a decision.

A clinical measurement review notes the scale's broad use and its limited descriptive anchors. Another review of CGI practice recommends condition-specific knowledge and clear reference points.

For repeat use, keep the target condition, evidence sources, and rater approach stable. If a different clinician rates the person, document that change.

Document the reason for each rating.

What the score cannot tell you

CGI cannot diagnose a disorder or identify why a person improved or worsened.

A rating is not a diagnosis.

It cannot show which symptoms changed. It also cannot separate symptom change from changes in daily function unless the clinician records that reasoning.

A CGI-I improvement rating does not prove that treatment caused the change. Natural recovery, other care, life events, measurement variation, and rater expectation can contribute.

The Efficacy Index does not replace a full adverse-effect assessment. A global category can miss a specific serious problem.

Do not use CGI alone to select treatment, judge safety, or close follow-up.

Keep the supporting clinical evidence with the number.

Reasoning matters.

Evidence and limitations

CGI is brief and easy to apply across conditions. It can summarize a clinician's overall judgment when disorder-specific scales do not capture the full picture.

Its main limitation is subjectivity. The original anchors leave room for different interpretations of the same clinical state.

Judgment varies.

Different clinicians may weigh symptoms, function, treatment expectations, and recent events differently even when they review the same person on the same day.

Knowledge of the person's history can improve the judgment but can also introduce expectation bias. Unblinded treatment studies are especially vulnerable.

Reliability improves when raters receive training, agree on the target condition and timeframe, and use the same information sources.

Calibration helps.

Report the component name, value, baseline when relevant, date, and rater. A CGI number without those details is not a complete clinical record.