Patient Health Questionnaire-9:
Scoring and Interpretation

The PHQ-9 records how often a person has experienced nine depressive symptoms during the previous two weeks. Kroenke, Spitzer and Williams (2001) evaluated it in US primary-care and obstetrics-gynecology settings as a brief measure of depressive symptom severity. Its content follows the diagnostic symptom criteria for depression, yet the summed score expresses symptom burden rather than a diagnosis.

Researchers use it in observational cohorts, behavioral interventions, clinical trials and patient-reported outcome studies, and the short format suits repeated measurement. In the original validation, higher scores were associated with poorer functioning and greater health-care use, and an individual participant data meta-analysis later widened the evidence on screening accuracy (Kroenke et al., 2001; Levis et al., 2019). Because the total is a plain sum, the scoring steps most often mishandled are missing items and the treatment of a screening threshold as a diagnosis.

• Items: nine self-reported symptom items that contribute equally to the total.
• Response scale: four-point frequency scale coded 0 to 3, referring to the previous two weeks.
• Subscales: none. The standard output is one total score.
• Reverse-scored items: none.
• Total: one summed score from 0 to 27.
• Unscored content: the difficulty or functional-impact question printed on some forms sits outside the 0 to 27 total and belongs in its own variable.
• Short forms: the PHQ-2 and the PHQ-8 (which omits the ninth item) exist and are not numerically equivalent to the PHQ-9.

1. Check that phq1 to phq9 contain integers from 0 to 3, and keep any functional-impact item in a separate variable.
2. Sum all nine responses to form phq9_total, which ranges from 0 to 27. No reverse coding and no standard-score transformation is needed.
3. Decide on a missing-item rule before analysis. The original validation does not prescribe one. A common convention prorates when one or two items are missing (the sum of answered items multiplied by 9 and divided by the number answered) and sets the total to missing when three or more are absent. Describe it as a convention in the methods section. The code below requires all nine items.

Higher totals mean more frequent depressive symptoms. For major depression at a cut-off of 10 or more, the original US validation reported sensitivity of 88% and specificity of 88%. The individual participant data meta-analysis by Levis et al. (2019) pooled sensitivity of 88% and specificity of 85% against semistructured interviews at the same threshold. A positive screen identifies someone who needs a clinical assessment; it is not a diagnosis.

Prespecify the PHQ-9 total as a continuous symptom score for descriptive summaries and for regression or mixed-effects models in which the distributional assumptions are reasonable, and reach for the published bands only when a categorization answers a clinical or reporting question. With repeated assessments the within-person structure has to be modeled: Linear Mixed Models (Multilevel Modeling) handle several waves and incomplete follow-up, while a single before-and-after comparison can use a Paired Samples t-Test reported as a mean change with its 95% confidence interval. State how missing data were handled and show the score distribution, including any floor effect.

Estimate reliability in your own sample, with a confidence interval when feasible; Reliability Analysis (Cronbach’s Alpha) covers that step. If the study aims to validate the measure or test its structure, a Confirmatory Factor Analysis (CFA) needs a sample size planned around model complexity and precision rather than a subjects-per-item rule. A minimal clinically important difference depends on context and population, so do not present a single figure as universal.