Shapes how you counsel, set expectations, or answer the question a patient brings in from the internet.
Front Health Serv·Cluster RCT, 20 clinicians, 587 patients·PMID 43016714·NCT05167227·open access·full text and registry record read; supplementary tables not opened
Fourteen authors from Family Health Centers of San Diego, San Diego State, the University of Washington, the University of Colorado, Ata Health Strategies, the ECHO Institute and the CDC — the last including Elizabeth Unger and Jeanne Bertolli — randomised 20 primary-care clinicians at a large federally qualified health centre to one of two levels of education in long COVID and ME/CFS. Both arms were offered monthly webinars and quarterly online courses. The intervention arm was also offered weekly Project ECHO case conferences, each with at least two national experts and a person with lived experience. Each clinician was assigned about 30 patients, 587 of whom completed a baseline survey (298 control, 289 intervention), and patients were followed by telephone survey for a year. The population was low-income (94.4% under $50,000 a year), 40% preferred Spanish, and 91.3% had long COVID alone; 36 had long COVID with ME/CFS and 13 ME/CFS alone. Two patients with severe ME/CFS could not complete the telephone survey at all.
What they report. The primary outcome, stated in the methods and in the sample-size calculation, was change in PROMIS-29 physical function. The treatment effect was 1.49 T-score points (95% CI −0.04 to 3.02; p = 0.06): no significant difference. Of the nine PROMIS scales modelled, only pain interference separated, by −1.95 points (−3.79 to −0.10; p = 0.04), and at 12 months the arms scored 57 against 60 from a shared baseline of 61. Among patients who completed both baseline and 12-month surveys, 43% of intervention patients (63 of 145) against 29% of controls (50 of 174) improved by at least 5 points on pain interference. Both arms improved on fatigue, physical function, social roles, dyspnoea and cognition over the year. Intervention patients received fewer referrals (median 9 against 10) and completed fewer encounters (16 against 21). No harms were reported to the review board.
Bottom lineThe only randomised evidence so far that training primary-care clinicians in this field helps patients, and it mostly shows that it did not change function. Report the primary outcome first: case conferences on top of webinars did not improve physical function. The pain signal is plausible and modest: about 1 in 7 more patients reached a meaningful improvement in how much pain interfered with life, which fits the idea that clinicians who learn from colleagues treat pain more actively. It comes from one of nine secondary scales with no correction for multiple comparisons, and the average effect is under half the authors' own 5-point threshold. Both arms improved substantially, which reflects natural recovery from long COVID in the first years after infection as much as anything either arm did, and is a reason not to read before-and-after improvement in an uncontrolled long COVID clinic report as a treatment effect. For her, the practical points are two. Peer case discussion is cheap and may help with pain. And this trial tells her almost nothing about ME/CFS specifically: 49 patients, with no analysis by diagnosis.
Weigh it accordingly. The abstract omits the primary outcome. It reports that both arms improved, then the pain-interference result, and concludes that the model “showed evidence” of benefit; the null physical-function result appears only in Table 3. The registry does not settle which outcome was primary: NCT05167227 lists the whole “PROMIS-29”, with no subscale, as its primary outcome, so the paper's choice of physical function cannot be checked against the registration, though it is consistent with the published protocol's sample-size reasoning as described. The registry also lists, as a secondary outcome, whether symptoms worsen for at least 24 hours after activity — a PEM question — which is not reported in the main text (it may be in the symptom-screener tables in the supplement, which this digest did not open). The discussion overstates its own results. It says the intervention arm “reported meaningful improvement in pain interference” when the results section says the average effect “might not be clinically meaningful”, and it describes the 43% and 29% responder proportions as though they were the two arms' scores. Fewer referrals and encounters are presented as a health-system benefit; they could equally mean patients received less care, and the paper has no way to tell the two apart. Retention was poor and unequal: 12-month surveys were completed by 50.5% of the intervention arm and 60.4% of controls (146 of 289; 180 of 298), and the authors say they may have lost both the sickest and the recovered. Patients were assigned to clinicians by clinic and language rather than at random; clinicians attended 42% of the case conferences offered; the control arm received active education, which the authors note may have shrunk the difference; and long COVID was defined as symptoms at four weeks, the definition current in 2021. The comparison with the University of Washington clinic is not randomised and its finding that San Diego patients did better says nothing about the intervention. Two small inconsistencies: the abstract says 590 patients “completed surveys” where 587 did; and Table 2 gives 180 and 146 completers at 12 months where the responder analysis uses 174 and 145. The registry's start date, 30 November 2021, precedes its first submission on 17 December; the paper says recruitment began in December. Interests: funded by the CDC, with CDC-affiliated authors, evaluating a programme the CDC paid for; one author was employed by a consultancy in the consortium. ChatGPT was used for spelling and grammar, declared. Arithmetic that checks: 298 + 289 = 587; 536, 36, 13 and 2 sum to 587; 63/145 = 43.4%; 50/174 = 28.7%; both treatment-effect intervals are symmetric about their estimates (1.49 ± 1.53; −1.95 −1.84/+1.85). Published 28 October, entered PubMed 12 November.
Brain Behav Immun Health·Retrospective chart review, single centre, no control group, n = 50·PMID 43032818·open access·full text read; supplementary tables not opened
Seventeen authors, most at the Stanford long COVID clinic, with Grace McComsey of Case Western, reviewed the records of the 76 of 1,134 clinic patients prescribed aripiprazole at 0.1–2 mg daily between May 2021 and September 2023 for long COVID symptoms. 50 had usable before-and-after questionnaires and form the analysis; anyone diagnosed with ME/CFS before their COVID infection was excluded. Follow-up was whenever the patient was next seen. Outcomes were a 29-symptom checklist the clinic designed itself, rated 0–5, and a five-level functional status scale adapted from the Post-COVID-19 Functional Status scale.
What they report. Functional level improved from a mean of 3.78 to 3.28 (p = 0.001): before treatment 8 patients were at level V (bedbound) and 23 at IV; afterwards 3 and 15, with 7 reaching level II. No symptom improved significantly — not fatigue, brain fog, unrefreshing sleep, headache, lightheadedness, nor post-exertional malaise (3.68 to 3.44, p = 0.34) — and neither did the symptom count or the composite score; 27 patients' composite improved, 22 worsened and 1 was unchanged. The median final dose was 1 mg; 18 reached 2 mg. Twelve of the 76 (15.8%) reported side effects. Weight changed in most: 25 gained up to 42 pounds and 18 lost up to 71. Most patients were also taking low-dose naltrexone. In a subgroup of 33 with two visits before and after, the authors report that the improvement in function persisted after allowing for the pre-treatment trend.
Bottom lineIf a patient asks about low-dose aripiprazole: the only data in long COVID are this chart review, in which patients said they were functioning a little better while reporting the same symptoms, with no comparison group, during the period when long COVID tends to improve on its own, and with most also on naltrexone. The authors themselves call for a randomised trial before it is used. Its PEM item did not move. Weight change and the stigma of an antipsychotic label are real costs, and one patient in six reported a side effect. The rationale was an earlier ME/CFS case series from some of the same clinicians, and that was uncontrolled too. It is reasonable to tell a patient that it is being tried, that it is not known to work, and that a trial is the right place to try it.
Weigh it accordingly. Single-centre, retrospective and uncontrolled, as the authors state, adding selection bias, concomitant treatments, uncontrolled follow-up duration, multiple exploratory comparisons and placebo effects to their own list of limitations. A third of those treated are missing (26 of 76) for incomplete records. Both outcome instruments are unvalidated: the methods say the functional scale was modified and “has not been independently validate[d]”, while the discussion calls it “a validated 5-point ordinal instrument”. Internal inconsistencies: the abstract describes symptom ratings from 1 to 5 where the body uses 0 to 5; the composite score is described as an average of seven symptoms divided by 7, but the table values (20.36 and 18.62) are exactly the sums of the seven symptom means; duration on treatment is given in consecutive sentences as an average of 315 days (range 31–1165) and a mean of 204.4 days (range 31–1025); the weight percentages (48.1%, 34.6%, 3.8%) and the 53.8% concomitant naltrexone figure correspond to a denominator of 52, not 50; and 43 of 50 on naltrexone is 86.0%, not the 86.6% printed. PubMed indexes the paper as a review, which it is not. Arithmetic that checks: the functional-level distributions sum to 50 before and after and give the stated means exactly (189/50 and 164/50); 27 + 22 + 1 = 50; 12/76 = 15.8%. No funding statement was found in the text read; the authors declare no conflicts. Ethics: Stanford IRB 62996, consent waived. Received 5 April, accepted 27 September, online 6 October, entered PubMed 14 November.
Acta Psychol (Amst)·Pilot RCT, 79 analysed (46 vs 33)·PMID 43025223·abstract and declarations only; full text not read
Read the meta line first. The paper is on ScienceDirect, which has served this digest bot checks on earlier runs, and has no PubMed Central copy; the full text was not attempted, so everything here comes from the MEDLINE abstract. Eighteen authors at Biobizkaia, Cruces, Donostia and Araba university hospitals in the Basque Country randomised people with post-COVID condition or ME/CFS to individualised home cognitive and motor training four times a week, adjusted weekly, or to standard care, for 16 weeks. 79 had both assessments: 46 intervention, 33 control. Outcomes were neuropsychological tests, self-reported symptoms and the six-minute walk.
What they report. No between-group difference in any comparison and no domain-level time-by-group interaction after false-discovery-rate correction. Within the intervention arm alone, sustained attention and semantic verbal fluency improved (both corrected p = 0.028). An exploratory component explaining 27.2% of cognitive change was higher in the intervention arm (β = 0.715, p = 0.043, uncorrected) and tracked motor adherence. Median recorded adherence was 99.75% for motor and 100% for cognitive tasks; median usability score 77.5. The authors' conclusion is restrained: feasibility is supported, differential benefit was not established.
Bottom lineAnother null for cognitive training in post-viral illness, worth having when a patient asks whether a brain-training programme will help her brain fog: in this randomised comparison, which included ME/CFS, it did no better than standard care. Within-arm improvement in an untreated-control design is what practice effects on repeated testing look like. The abstract says nothing about whether training four times a week provoked post-exertional symptoms, although the background names them, and that is the question she would most want answered before suggesting it.
Weigh it accordingly. Abstract only. The number randomised, the case definitions, how many had ME/CFS rather than post-COVID condition, severity, dropouts and whether any adverse events were recorded are all unknown. 46 against 33 analysed suggests either unequal allocation or more loss from the control arm; the abstract does not say which. Adherence of 99.75–100% is platform-recorded and implausibly near-perfect for a fatiguing illness; the authors note that the clinical support behind it was not recorded. The one positive between-group signal is an uncorrected exploratory component. No funder appears in the record; the authors report no conflicts. Published and entered PubMed 13 November.