Two items from the history against a battery of eight instruments
- Cross-sectional study at a specialised drug rehabilitation centre in Hunan province, China, run by the Second Xiangya Hospital of Central South University in Changsha and published open access in BMC Psychiatry on 23 May 2026. Data were collected between March and December 2024 from 122 male patients aged 18 to 35, mean age 23.13 ± 6.71 years, mean education 9.70 ± 2.09 years. Entry required etomidate use at least twice a month for three months or more within the past year, confirmed by hair toxicology; polydrug use and current or past schizophrenia, major depressive, anxiety or bipolar disorder were exclusions.
- Antisocial personality disorder was diagnosed with MINI 7.0, the structured interview keyed to DSM-5, administered by two psychiatrists working in addiction medicine, with a Cohen's kappa of 0.86 measured between them before the study began. Sixty of the 122 patients, 49.2%, met the criteria. Across the whole sample 58 patients, 47.5%, had a criminal record and 35, 28.7%, reported a traumatic event.
- Age, education, marital status, employment, income and family structure returned p > 0.05 between the two groups. A criminal record separated them at p = 0.019 and a traumatic event at p = 0.007. Beck Anxiety Inventory scores were 11.45 ± 10.56 against 7.55 ± 10.05 (p = 0.039), Beck Depression Inventory 11.65 ± 6.53 against 7.40 ± 5.43 (p < 0.001), Barratt Impulsiveness Scale 48.24 ± 15.06 against 38.59 ± 11.05 (p < 0.001).
- Five variables entered the final logistic model after selection by the Boruta algorithm and LASSO regression: criminal record, odds ratio 3.48 (95% CI 1.40 to 8.67); traumatic event, 3.08 (1.12 to 8.48); craving, 1.52 (1.08 to 2.15) per point; attentional impulsiveness, 1.06 (1.03 to 1.09) per point; physical aggression, 1.04 (1.02 to 1.07) per point. The area under the curve was 0.840 (95% CI 0.770 to 0.909, p < 0.001), internally validated by bootstrap resampling over 1,000 repetitions and calibrated against the Hosmer-Lemeshow test.
What changes in the plan for a 23-year-old man admitted for etomidate use once he also meets the criteria for antisocial personality disorder? In one rehabilitation centre in Hunan province that question arrives for one patient in two, and the study behind it spends its effort on a narrower one: which parts of the intake carry the signal.
The Hunan sample
Recruitment ran from March to December 2024 and produced 122 male patients between 18 and 35, all of them using etomidate, a drug usually inhaled after being mixed into a conventional or electronic cigarette, all confirmed by hair toxicology. Comorbid psychiatric diagnoses were screened out at entry, so the sample holds one substance and, for half of it, one personality disorder. The men carried 9.70 ± 2.09 years of education, half were unmarried, 74.6% were in work and 57.4% reported a monthly income below 10,000 RMB; a quarter, 25.4%, came from single-parent families. Both diagnoses came from MINI 7.0 administered face to face by two addiction psychiatrists, not from a self-report scale, and the pair agreed at a kappa of 0.86 before any patient was enrolled. Sixty patients, 49.2%, met the DSM-5 criteria.
Two history items against three scales
The battery ran to eight instruments – craving, impulsivity, aggression, anxiety, depression, perceived stress, life satisfaction, social support – alongside two lines of the history: a criminal record and a traumatic event. Taken one at a time, most of the battery did tell the groups apart. Anxiety scores stood at 11.45 ± 10.56 against 7.55 ± 10.05 (p = 0.039), depression at 11.65 ± 6.53 against 7.40 ± 5.43 (p < 0.001), total impulsivity at 48.24 ± 15.06 against 38.59 ± 11.05 (p < 0.001), and perceived stress differed on the uncertainty subscale (p = 0.001) but not on tension. Selection by the Boruta algorithm and LASSO regression left five variables, and the odds ratios attached to them are of two different sizes. The criminal record carries 3.48 (95% CI 1.40 to 8.67) and the traumatic event 3.08 (1.12 to 8.48). Craving carries 1.52 (1.08 to 2.15) for each point of the desire questionnaire, attentional impulsiveness 1.06 (1.03 to 1.09) per point and physical aggression 1.04 (1.02 to 1.07) per point. Those last two run across scales scored in tens of points, so their contribution to the total accumulates rather than arriving in one step. The model reached an area under the curve of 0.840 (95% CI 0.770 to 0.909, p < 0.001) with bootstrap internal validation over 1,000 repetitions.
The bounds the authors set
Neither history item is rare in this ward. A criminal record appears in 47.5% of the sample and a traumatic event in 28.7%, and 62 men passed through the same filters without the diagnosis, so two questions sort the population coarsely rather than settling it. The authors state that their model could be referenced clinically after external validation and improvement, and they had none: validation was internal, on resamples of the same 122 patients. They list the sample as male only, drawn from a single region, cross-sectional, dependent on self-report for the scale variables, and built from risk factors with no protective factors in it.
A criminal record carried an odds ratio of 3.48 and a traumatic event 3.08, while three scored questionnaires carried 1.52, 1.06 and 1.04 for each point on their scales.
All 122 participants were male, recruited at one centre in one province, and the design is cross-sectional, so nothing here separates what preceded what. Validation was internal only, by bootstrap resampling of the same patients, and the authors say the model needs external validation before clinical reference. The scale variables come from self-report questionnaires administered inside a rehabilitation centre. Diagnosis rests on MINI 7.0, a structured screening interview rather than a full personality assessment, and no dimensional description of severity or trait domains was collected, so the study speaks to a categorical DSM-5 label and to nothing in the ICD-11 model. Age of onset, duration of etomidate use, length of abstinence and somatic symptoms were not analysed, and the authors name those as the gaps to fill next. The exclusion of polydrug use and of schizophrenia, depressive, anxiety and bipolar disorder removes the comorbidity that most patients in an addiction service carry, so the 49.2% figure describes an unusually clean sample. Protective factors were not entered into the model. No treatment was given, tested or followed, and the study reports no outcome beyond the presence of the diagnosis.