Design: secondary, item-level path analysis of a randomised controlled trial
Population: 216 primary care patients, mean age 59, 78% women, 50% Black, 46% with income under $10,000 a year
Primary outcome: item-level effects on the SCL-20 and PHQ-9 at 12 months
Effect: better than usual care on 15 of 20 SCL-20 and 7 of 9 PHQ-9 items; 17 symptoms with β −0.50 to −0.70; no advantage for suicidality, loneliness, sexual dysfunction, appetite or psychomotor symptoms (β −0.03 to −0.37)

Trials of depression treatment report a total score, and a total score can fall while the symptom that worries you most does not. This secondary analysis of the eIMPACT trial modelled item-level effects in 216 primary care patients, half of them Black and 46% with an income under $10,000 a year, randomised to twelve months of modernised collaborative care — internet or telephone cognitive behavioural therapy, selected antidepressants, or both — against usual primary care.

The intervention beat usual care on 15 of 20 items on the Hopkins Symptom Checklist and 7 of 9 on the PHQ-9. Seventeen symptoms improved moderately to largely, with standardised coefficients between −0.50 and −0.70: hopelessness, anxious distress, depressed mood, self-blame, worthlessness, disturbed sleep and fatigue among them. Six more improved modestly. Five showed no significant advantage at all — suicidality, loneliness, sexual dysfunction, appetite change and psychomotor symptoms, with coefficients from −0.03 to −0.37.

That list is the finding. Suicidality and loneliness are exactly the domains a clinician is watching for, and exactly the ones a falling PHQ-9 total will conceal. If you run or refer into a collaborative care programme, the implication is not to stop — it is to track those five separately, and to treat a good total score as the beginning of the review rather than the end of it.

Read PHQ-9 item 9 on its own at every review, not as part of the total.
Ask about loneliness explicitly; it does not appear on the scales most services use.
Ask about sexual dysfunction directly — the patient will rarely raise it and the antidepressant may be causing it.
Record weight and appetite over time; neither moved with treatment here.
Treat a responding total score with persistent psychomotor slowing as an indication for review, not discharge.

Why it matters

A falling total score is the usual signal to step down review intensity, and this says the total hides the symptoms you would step down on.

The statistics, in plain English

The coefficients here are standardised between-group differences, so −0.70 is a large effect and −0.03 is effectively nothing. "No significant advantage" for suicidality with 216 patients means the trial could not show a benefit, not that it proved the absence of one — suicidality is uncommon enough that a study this size has limited power to detect change in it. Either way, no clinician should assume the item is covered.