Clinical Global Impression (CGI-S & CGI-I)

The clinician answers one question: "Considering your total clinical experience with this particular population, how mentally ill is the patient at this time?" Rated on a 7-point scale: 1 = normal (not at all ill) to 7 = among the most extremely ill patients. The rating is based on observed and reported symptoms, behavior, and function over the past 7 days — it is a state-dependent measure summarizing one week, not a lifetime of symptoms (Busner & Targum, 2007). Because the anchor depends on the clinician's experience with the specific patient population, a score of 4 (moderately ill) may represent very different symptom burdens across disorders.

Rated 1 (very much improved) to 7 (very much worse) compared to the patient's condition at the start of treatment (baseline). The CGI-I does not ask whether change is due to medication — it captures the clinician's global judgment of change regardless of attributed cause. This means side effects unrelated to the target condition should not influence the rating, though in practice they often do (Busner et al., 2009). The CGI-I generally tracks with the CGI-S, but the two can dissociate: a patient may show improvement on the CGI-I relative to baseline despite no recent change in the CGI-S, or vice versa.

No universally accepted anchor guidelines exist for the seven points — they were deliberately designed to rely on clinical judgment. Busner & Targum (2007) published generically applicable guidelines suggesting, for example, that CGI-S 4 (moderately ill) reflects overt symptoms causing noticeable but modest functional impairment, while CGI-S 6 (severely ill) reflects disruptive pathology where behavior and function are frequently influenced by symptoms and the patient may require assistance from others. These are suggestions, not absolutes — clinical gestalt may warrant scoring above or below what the anchors suggest.

Inter-rater reliability for the CGI varies substantially by diagnosis, domain, and rater training. In the CGI-SCA (schizoaffective disorder) development study, Allen et al. (2012) reported ICCs ranging from 0.50 (overall severity) to 0.88 (depressive domain) across 12 videotaped interviews rated by two independent clinicians. The relatively low ICC for the overall rating reflected a specific problem: some raters used the recommended gestalt method while others used an arithmetic average of domain scores — a distinction that training materials must address explicitly. In contrast, the CGI-SCH (schizophrenia) study by Haro et al. (2003) found ICCs above 0.70 for all domains except depressive (0.64), and the CGI-AS (Angelman syndrome) adaptation by Kolevzon et al. (2021) achieved ICCs of 0.85–0.99 across six domains after rigorous rater training with standardized video vignettes. The takeaway: the CGI is only as reliable as the training behind it.

A recurring difficulty across all multi-domain CGI adaptations is distinguishing negative from depressive symptoms. In the CGI-SCA study, raters anticipated this problem during content validity surveys (74.8–86.2% agreement depending on region), and the data confirmed it: the CGI-SCA negative and depressive domain scores correlated at r = 0.40 at baseline, and the CGI-SCA depressive domain correlated with the PANSS negative factor at r = 0.318 — nearly as high as with depression-specific measures (Allen et al., 2012). A similar pattern emerged in the CGI-SCH (Haro et al., 2003), where the depressive domain had the lowest ICC (0.64). In clinical practice, this means that a patient whose negative symptoms are misattributed to depression — or vice versa — may receive an inaccurate domain score, even if the overall CGI remains valid.

Busner et al. (2009) demonstrated experimentally that CGI ratings are susceptible to a specific bias that most clinicians do not recognize. They randomized 45 principal investigators to receive identical patient narratives with or without added medical or psychiatric adverse events unrelated to the condition under study. The result: CGI-S and CGI-I scores were significantly worse in the adverse-event condition (P < .04). Yet 46% of investigators reported that psychiatric adverse events would not affect their CGI-S ratings, and 87% said medical adverse events would not — despite the data showing otherwise (P < .03 for CGI-S). This gap between perceived and actual susceptibility means clinicians need a conscious mental checklist to exclude non-target symptoms from the global rating.

The Kolevzon et al. (2021) Angelman syndrome CGI development provides one of the most detailed explorations of this question. Through structured interviews with 9 clinical experts and 22 caregivers, they established that a CGI-I rating of 3 (minimally improved) represents the threshold for minimally clinically meaningful change — but only when the improvement is seen consistently in at least one setting and represents more than random fluctuation. A CGI-I of 2 (much improved) requires improvement across settings, noticed by multiple observers. The caregivers consistently anchored meaningfulness to functional impact: improved communication reduced frustration and behavior problems; better sleep reduced caregiver fatigue and marital tension; improved gross motor function enabled community outings. These domain-specific anchors are now built into the CGI-AS rater training program and illustrate how CGI-I ratings can be grounded in observable, functional endpoints rather than abstract severity shifts.

Berk et al. (2008) tracked CGI completion in 786 consecutive psychiatric admissions over 24 months. CGI-S was completed in 79.4% of admissions and 78.1% of discharges; CGI-I in 77.6%. These rates were comparable to the Health of the Nation Outcome Scales (HoNOS) and superior to the Depression Anxiety Stress Scales (DASS-21), suggesting that the CGI's brevity translates into better real-world data capture. The correlation between admission and discharge CGI-S was r = 0.40, and the indirect improvement measure (admission CGI-S minus discharge CGI-S) correlated strongly with the direct CGI-I at r = 0.71, supporting the validity of both measurement approaches.

The CGI has known blind spots. It does not capture functional outcomes directly — a patient may show symptom improvement on the CGI-I while remaining severely functionally impaired. It is vulnerable to ceiling effects in milder illness (most patients cluster at scores 2–4, limiting discrimination). And because it is a single-item global rating, it cannot distinguish between improvement in one symptom domain and worsening in another. The Allen et al. (2012) CGI-SCA data illustrate this: the overall severity score correlated most strongly with positive symptoms (r = 0.72) and manic symptoms (r = 0.40), but only weakly with negative (r = 0.24) and depressive (r = 0.20) symptoms. This means a patient whose positive symptoms improve but whose negative symptoms worsen may appear unchanged on the global CGI — a problem that domain-level CGI versions partly address.

The Clinical Global Impression (CGI) scale was developed by the NIMH Psychopharmacology Research Branch in the early 1970s as part of the Early Clinical Drug Evaluation Program (ECDEU), led by William Guy. The goal was pragmatic: create a brief, clinician-rated global assessment that could be used across all psychiatric diagnoses in multicenter drug trials. The original manual (Guy, 1976) described the CGI as a two-component instrument — severity and improvement — each rated on a 7-point scale. Despite — or perhaps because of — its simplicity, the CGI has proven remarkably durable. It is now the most widely cited global outcome measure in psychiatry, included in virtually every FDA-regulated CNS clinical trial over the past four decades. Its adaptability has spawned condition-specific versions for bipolar disorder, schizophrenia, schizoaffective disorder, Alzheimer's disease, borderline personality disorder, social anxiety disorder, and most recently, rare neurodevelopmental conditions such as Angelman syndrome and Rett syndrome.