GDS-15 (Geriatric Depression Scale): scoring and interpretation

The 15-item Geriatric Depression Scale has an unusual problem: it is free, and it is still hard to know which copy you are holding. Different official-looking sources publish different cutoffs, and at least one national clinical guideline prints an item that is not on the validated form. This guide covers how it is scored, why the thresholds disagree, where it stops working, and what its public-domain status actually covers.

What it measures, and why a scale only for older adults

The original 30-item scale was built from a 100-question pool, keeping the items most correlated with the total and validating them against Research Diagnostic Criteria (Yesavage et al., 1983). The 15-item version is a derivative Sheikh and Yesavage published in 1986 as a book chapter that is not open access, so what is publicly known about its cutoff comes from the rights holder’s page, later validations and the guidelines that adopted it.

The design rationale matters more than the history: the GDS asks about no somatic symptoms. The test review by Spain’s Consejo General de la Psicología puts it plainly — the GDS contains no somatic complaints that may be more common in older people, sleep disturbance among them. None of the 15 items asks about sleep, appetite, weight or pain. In a 78-year-old with diabetes, arthritis and insomnia, the somatic items of a general instrument score the illness rather than the mood. The PHQ-9 and the GDS-15 are not rivals: the first grades severity across a wider range and carries a suicidal-ideation item, the second avoids that contamination entirely.

Administration and scoring

The recall window is one week, not two. The rights holder’s own form asks the respondent to choose the best answer for how they have felt over the past week.

Each item is answered yes or no and scores 1 or 0; administration takes about five minutes. Reading it aloud is not a deviation: in the Spanish validation a doctor or nurse administered it in every case, because 65% of its sample was non-literate or barely able to read and write. The maximum score is 15. Ten items indicate depressive symptoms when answered yes; five when answered no, and those five are items 1, 5, 7, 11 and 13 — life satisfaction, good spirits, happiness, feeling it is wonderful to be alive, feeling full of energy. Adding up every “yes” produces a number that means nothing.

The published cutoffs do not agree

Four legitimate sources, two thresholds.

Source	Positive from	Severity bands	Item 12
Yesavage / Stanford	more than 5 points	more than 10, “almost always depression”	worthlessness
INGER, Mexico’s Ministry of Health (2020)	5 or more	none: 0–4 normal	worthlessness
Mexican CPG IMSS-194-08 (2011)	6 affirmative answers	0–5, 6–9, more than 10	difficulty starting new projects
GDS-VE, Spain (2002)	5 or more, empirically derived	none	worthlessness

The rights holder writes that “a score > 5 points is suggestive of depression and should warrent a follow-up interview”, which makes the screen positive at 6. Most downstream literature reports “cutoff 5”, meaning 5 or more. That single point changes who gets flagged.

Mexico is the clearest illustration, because two federal sources disagree with each other. The Instituto Nacional de Geriatría gives two categories only: normal at 0–4, depressive symptoms present at 5 or more. The clinical practice guideline IMSS-194-08 states the scale is positive when 6 answers are affirmative. A clinician following one national document will flag patients the other leaves in the normal band.

That guideline shows two further things. Its printed annex gives item 12 as “¿Le cuesta iniciar nuevos proyectos?” — do you find it hard to start new projects — where the standard item 12 is the worthlessness question, and the projects item belongs to the 30-item scale; the annex is nonetheless credited as adapted from Sheikh and Yesavage, 1986. And the 97% sensitivity and 85% specificity it attributes to the 30- and 15-item GDS are the figures Hoyl and colleagues (1999) reported for the 5-item GDS.

Do not resolve this by moving the threshold case by case: pick a version and a criterion, write it down, and apply it the same way every time. In Mexico the strongest case is INGER’s: it is official, and its 15 items are the validated Spanish version’s, bar gender spellings.

The severity bands are a separate problem. Stanford does publish bands, but for the 30-item scale only: normal 0–9, mild 10–19, severe 20–30. For the 15-item form we found none, and the band schemes in wide circulation — including the one in the Mexican guideline annex — could not be traced to any rights-holder publication or to the original papers. If your report prints “moderate depression” from a GDS-15, that adjective lacks the authority it appears to have.

How well it screens

The Spanish adaptation (GDS-VE) is a validation rather than a translation: 249 people over 65 in a primary-care centre in Córdoba, with a second investigator conducting a blind semi-structured ICD-10 interview as the gold standard. Area under the curve was 0.835, and the grid lets you pick a threshold deliberately:

Cutoff	Sensitivity	Specificity	Misclassified
4	86.7%	61.6%	29.3%
5	81.1%	76.7%	21.7%
6	72.6%	85.5%	20.9%
7	51.1%	90.6%	23.7%

At the conventional cutoff of 5, pooled sensitivity was 0.89 (95% CI 0.80–0.94) and specificity 0.77 (0.65–0.86), with a caveat from the authors themselves: possible selective reporting of cutoff scores (Pocklington et al., 2016).

What a score does not tell you

Cognitive impairment is a hard floor

In the Spanish validation, area under the curve was 0.865 without cognitive impairment (n=192) and 0.745 with it (n=57). In 66 nursing-home residents the scale returned 63% sensitivity and 83% specificity, and in the 44 with an MMSE of 15 or above that rose to 84% and 91%: screening cognition first significantly increases its utility (McGivney et al., 1994: cutoff 10 or more, version not specified). In mild cognitive impairment it holds up; in Alzheimer’s disease it does not — against the Cornell Scale for Depression in Dementia, sensitivity and specificity were too low to select an optimal cutoff by ROC analysis at all (Debruyne et al., 2009, using the 30-item form).

The mechanism is not the one usually assumed. Patients with Alzheimer’s who disavow their cognitive deficits also tend to disavow depressive symptoms, and worse memory was associated with more self-reported symptoms, not fewer (Feher et al., 1992). Assess cognition first; with established dementia, a low GDS-15 is not reassuring.

There is no suicide item

None of the 15 items asks about suicidal ideation or self-harm. Item 14 asks about hopelessness, which is not the same thing. A GDS-15 of 2 does not rule out suicide risk, and unlike the PHQ-9 there is no item here that forces you to look.

Any risk disclosure calls for an immediate clinical risk assessment under your own protocol and training. This guide will not hand you a step-by-step protocol — that comes from the professional guidelines that apply where you practise. What you can settle in advance is how that assessment gets documented, and that the client leaves knowing whom to contact between sessions: the crisis line serving your client’s location, plus local emergency services.

It is not a diagnosis, and it is a short ruler

The IMSS guideline’s glossary says the instrument does not replace professional clinical assessment, and INGER heads its thresholds as “suggestions or guidance for interpretation” rather than rules. The range, 0 to 15 in whole-point steps against the PHQ-9’s 0 to 27, also makes it blunt for tracking progress session by session — a limit of resolution, not a documented floor or ceiling effect.

The Spanish-language evidence, and the gap in it

Translation is not validation, and reliability is not diagnostic accuracy — a distinction that matters if you screen Spanish-speaking older adults. Mexico has recent psychometric evidence: a study of 1,178 people aged 60 to 94 in Sonora reported internal consistency of .80 and divergent validity, with an instrument translated and back-translated for that study (Acosta Quiroz et al., 2021); a second, with 150 older adults, found a unidimensional structure and ordinal alpha of .94 (Salgado Cedano et al., 2024). Neither used a clinical interview as a criterion or derived a cutoff.

We could not locate any Mexican diagnostic-accuracy study comparing the GDS-15 against a structured clinical interview and deriving a local threshold, so “the GDS-15 is validated in Mexico” repeats an error that is everywhere online. The only accuracy data in people of Mexican origin is old, small and unflattering: in 41 psychiatric inpatients, sensitivity at the original cut-scores was 39% for major depression, and with no true negatives specificity could not be computed (Baker & Espino, 1997). A flag, not a verdict.

Licensing: what is public domain, and what exactly

Read the rights holder’s sentence precisely, because it does not say what it is usually quoted as saying: “The original scale is in the public domain due to it being partly the result of Federal support.” That is the original scale, the 30-item one; the 15-item form is a 1986 derivative. The rights holder nevertheless distributes the 15-item form and its scoring key itself, along with free apps and a web page for administering it and copying the result into a record — an affirmative offer of free clinical use, and that is the permission a clinician needs. Write it that way, not as “the GDS-15 is in the public domain”. Of the translations it hosts, it warns it cannot vouch for their accuracy.

Translations carry their own terms. Mexico’s federal Spanish version is a case in point: one PDF, three statements that do not agree — the legal page declares all rights reserved; the credits declare a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International licence; and the instrument annex permits reproducing and distributing it “in its entirety, for educational and/or care purposes only, not for profit, provided the Instituto Nacional de Geriatría is cited as the source”. Whether a fee-charging practice falls inside “not for profit” is genuinely unclear, and we found no clarification. Keep the official PDF with its date, cite the source, and ask of any instrument you use: who holds the rights, what their page says, and which edition and language the terms cover.

Where the score lives in the note

Record-keeping rules rarely name instruments, but they say what an entry must contain and who signs it. Mexico’s clinical-record standard is a workable example: progress notes must carry the relevant results of previously requested diagnostic and treatment studies, and every note a date, a time, the author’s full name and a signature. Verify the equivalent rule where you practise.

“Administered the 15-item Geriatric Depression Scale (GDS-15), read aloud, one-week recall window. Score: 8 of 15; this version treats 5 or more as depressive symptoms present. Screening result, not a diagnosis; corroborated against clinical interview this session.”

The score belongs in the O section of a SOAP note and your reading of it in the A; recording which version and criterion you used is what makes today’s number comparable to the one three months from now.

Choosing a version, scoring the reverse-keyed items correctly and comparing against the last measurement is easy with one client and heavy with thirty. That is where a structured chart helps: gesell.ai includes built-in validated scales with automatic scoring, PHQ-9 and GAD-7 among them, integrated into each client’s chart, so the scores you record sit beside the previous ones, ready for your note. Always as decision support you evaluate: the scale contributes the data point, the clinical judgment stays yours.