This analytical cross-sectional study was conducted at a specialized outpatient HIV clinic in Querétaro, Mexico. The analytic dataset comprised adults receiving care between January and December 2021. Cognitive assessment was performed prospectively during the study period after enrollment and written informed consent, whereas clinical and laboratory variables were obtained from the available medical records. This report was prepared in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement [8].

Eligible participants were 18 years or older, had confirmed HIV infection, and provided written informed consent. Exclusion criteria were a recorded history of neurological or systemic conditions considered likely to independently and substantially affect cognitive testing, including cerebral toxoplasmosis, cryptococcal meningitis, central nervous system lymphoma, HIV encephalopathy, progressive multifocal leukoencephalopathy, previous stroke, severe hearing impairment, or complete blindness.

Participants were enrolled consecutively as they attended outpatient consultations during the study period, provided that they met the eligibility criteria and agreed to participate. No random selection was performed; the recruitment strategy was therefore consecutive non-probability sampling. All enrolled individuals in the study database who met the eligibility criteria and had a recorded MoCA score were included. The database did not contain the total number approached or screened, the number who declined participation, or detailed reasons for non-enrollment; consequently, a participation fraction and complete recruitment flow diagram could not be reconstructed. The available analytic sample determined the study size, and no formal sample-size calculation was performed.

Cognitive screening was performed using the Mexican Spanish-language MoCA [6][9]. Each participant underwent a single MoCA assessment after providing written informed consent and enrolling in the study. No participant underwent more than one MoCA assessment during the study period; therefore, no selection among repeated assessments was required. The primary outcome was the total MoCA score obtained during the study assessment, ranging from 0 to 30, analyzed as a continuous variable. The dataset contained the final total score but not item-level responses, examiner certification information, or documentation confirming whether the standard educational correction had been applied. No retrospective score correction was performed.

For comparison with the original protocol and previous literature, the number of participants with a score below 26 was also summarized. This threshold analysis was explicitly descriptive: a score below 26 was not labeled neurocognitive impairment, was not interpreted as a diagnosis of HAND, and was not used to estimate the prevalence of a pathological condition.

Available variables included age, sex, schooling categorized as more than 12 years versus 12 years or less, occupation, time since HIV diagnosis, body mass index, hemoglobin concentration, recorded comorbidities, antiretroviral regimen class, current plasma HIV viral load, CD4⁺ T-cell count at diagnosis, current CD4⁺ T-cell count, and the clinical-immunological category documented at initial HIV diagnosis.

The clinical-immunological category used in this study was the category documented at the time of the initial HIV diagnosis and was analyzed as a baseline historical characteristic rather than current disease status; it was not reconstructed from subsequent clinical status. AIDS at initial diagnosis was defined from this recorded category as A3, B3, C1, C2, or C3. Categories A1, A2, B1, and B2 were classified as not meeting this definition. Records stating that the initial category was unknown were retained as missing in the primary analysis. For laboratory variables, the most recent value available at the time of the MoCA assessment was used, and no laboratory value was more than 3 months old. Current viral load was classified as greater than 50 copies/mL versus 50 copies/mL or less/undetectable when interpretable. One CD4⁺ T-cell value at diagnosis of 212,349 cells/mm³ was considered biologically implausible and treated as missing.

Comorbidity status was abstracted from the medical record. Recorded conditions included latent tuberculosis, hepatitis B or C, hypertension, diabetes, chronic kidney disease, asthma, hypothyroidism, schizophrenia, ankylosing spondylitis, anal cancer, and lymphoma; several entries indicated a comorbidity without specifying the diagnosis. Depression, anxiety, sleep disorders, active substance use, syphilis, and other potential cognitive confounders were not systematically measured with standardized instruments.

Categorical variables were summarized as counts and percentages using the number with available data as the denominator. Continuous variables were summarized as medians and interquartile ranges because several distributions were non-normal.

The primary inferential analysis modeled total MoCA score as a continuous outcome using ordinary least-squares linear regression with HC3 heteroscedasticity-consistent standard errors. Age, schooling of 12 years or less, and AIDS at initial diagnosis were selected a priori rather than through significance testing. Age and education were included because they are established determinants of MoCA performance, and AIDS at diagnosis represented the principal HIV-related hypothesis. The primary model used complete cases for these variables. Missing values were not imputed.

To evaluate whether complete-case exclusion might have influenced the primary-model estimates, participants with known versus unknown AIDS status at initial diagnosis were compared on age and MoCA score using Mann-Whitney U tests and on schooling category using Fisher’s exact test.

Sensitivity analyses included 1) a model retaining unknown AIDS status as a separate category to include all participants, 2) median regression because the MoCA distribution was bounded and left-skewed, and 3) logistic regression using the historical score threshold below 26 solely for comparison with the original analysis. The functional form of age was additionally assessed using a natural cubic spline with 3 degrees of freedom and by comparing the spline specification with the linear age specification. Model diagnostics included variance inflation factors, the Breusch-Pagan test, Ramsey’s regression specification error test, residual inspection, and Cook’s distance. For the influential-observation sensitivity analysis, the primary model was refitted after excluding observations with Cook’s distance greater than 4/n. Results are reported as beta coefficients or odds ratios with 95% confidence intervals. Two-sided P values below 0.05 were considered statistically significant, but interpretation emphasized effect estimates and confidence intervals. Analyses were performed using Python 3.13, pandas 2.2.3, and statsmodels 0.14.6.

The study was conducted in accordance with the Declaration of Helsinki and applicable Mexican regulations for health research. The relevant institutional bioethics committee approved the protocol, and all participants provided written informed consent before enrollment. Identifying committee details and the approval number are provided on the separate title page.

A total of 242 participants were included. Median age was 31 years (interquartile range [IQR], 28 - 38), 233 participants (96.3%) were male, and 207 (85.5%) had more than 12 years of schooling. Median time since HIV diagnosis was 4 years (IQR, 2 - 7). Among participants with an interpretable current viral load, 216 of 229 (94.3%) had 50 copies/mL or less or an undetectable result. The median current CD4⁺ T-cell count was 518 cells/mm3 (IQR, 356 - 714; n = 200), and 240 participants (99.2%) were receiving an integrase strand transfer inhibitor-based regimen.

Initial AIDS status was unavailable for 51 participants (21.1%). Among the 191 participants with a known initial category, 64 (33.5%) met the study definition of AIDS at diagnosis. CD4⁺ T-cell count at diagnosis was available and plausible for 170 participants, current CD4⁺ T-cell count for 200, current viral-load status for 229, and hemoglobin for 239. Table 1 summarizes the cohort and analytic denominators.

Participants with unknown AIDS status at initial diagnosis (n = 51) were slightly older than those with known status (median, 34 [IQR, 28 - 40] vs 31 [IQR, 28 - 37] years; P = 0.045). MoCA scores were similar between groups (median, 25 [IQR, 24 - 27] vs 26 [IQR, 24 - 28]; P = 0.750), as was the proportion with 12 years or less of schooling (6/51 [11.8%] vs 29/191 [15.2%]; P = 0.657).

Table 1. Characteristics of the study population.

Values are median (interquartile range) or n (%). Percentages use available data as the denominator. The score < 26 threshold is descriptive and does not define a neurocognitive disorder. One biologically implausible CD4⁺ T-cell count at diagnosis was set to missing. ART indicates antiretroviral therapy; INSTI, integrase strand transfer inhibitor; MoCA, Montreal Cognitive Assessment.

The median MoCA score obtained during the study assessment was 26 (IQR, 24 - 28), the mean was 25.6 (standard deviation, 2.7), and scores ranged from 15 to 30. A total of 114 participants (47.1%) scored below 26. This proportion describes performance relative to a historical threshold and does not represent the prevalence of HAND or another clinical neurocognitive disorder.

In univariable linear regression, older age was associated with a modestly lower MoCA score (beta = −0.047 points per year; 95% CI, −0.087 to −0.008; P = 0.019). Schooling of 12 years or less was associated with an estimated 0.95-point lower score, although the confidence interval included no difference (beta = −0.948; 95% CI, −2.007 to 0.111; P = 0.079). AIDS at initial diagnosis was associated with an estimated 0.56-point lower score in the subset with known initial category (beta = −0.560; 95% CI, −1.395 to 0.274; P = 0.188). No clear univariable association was observed for sex, duration of HIV, recorded comorbidity, body mass index, hemoglobin, or current viral load greater than 50 copies/mL (Table 2).

Table 2. Univariable and primary multivariable linear regression of total MoCA score.

Estimates are beta coefficients (95% confidence intervals) and represent the mean difference in MoCA points per unit or category. The adjusted model included age, schooling ≤ 12 years, and AIDS at initial diagnosis and used complete cases (n = 191). HC3 robust standard errors were used. BMI indicates body mass index; MoCA, Montreal Cognitive Assessment.

The primary multivariable model included 191 complete cases. After adjustment, age (beta = −0.020; 95% CI, −0.071 to 0.032; P = 0.458), schooling of 12 years or less (beta = −0.687; 95% CI, −1.963 to 0.589; P = 0.291), and AIDS at initial diagnosis (beta = −0.445; 95% CI, −1.306 to 0.417; P = 0.312) were not clearly associated with MoCA score. The model explained little outcome variation (R² = 0.020; adjusted R² = 0.005). Predictor variance inflation factors were low (maximum, 1.06), and there was no evidence of heteroscedasticity by the Breusch-Pagan test (P = 0.551) or major specification error by Ramsey’s test (P = 0.237).

In the sensitivity model retaining unknown AIDS status as a category (n = 242), age was associated with a small decrease in score (beta = −0.045 per year; 95% CI, −0.086 to −0.004; P = 0.031), whereas schooling of 12 years or less (beta = −0.917; 95% CI, −2.011 to 0.177; P = 0.100), AIDS at diagnosis (beta = −0.331; 95% CI, −1.183 to 0.522; P = 0.447), and unknown AIDS status (beta = 0.022; 95% CI, −0.789 to 0.832; P = 0.958) were not clearly associated (Table 3). The age estimate

Table 3. Sensitivity analysis retaining unknown AIDS status as a separate category (n = 242).

The reference category for AIDS status was no AIDS at initial diagnosis. HC3 robust standard errors were used.

in the full-sample model was modestly more negative than in the complete-case model (beta = −0.020; 95% CI, −0.071 to 0.032), and the confidence intervals overlapped substantially.

A natural cubic spline with 3 degrees of freedom did not improve model fit compared with a linear age specification in the complete-case model (P = 0.846 for the nonlinear component) or the full-sample model (P = 0.853). Descriptively, mean MoCA scores declined across age groups: 26.13 at 18 - 29 years, 25.39 at 30 - 39 years, and 24.73 at 40 years or older. These findings supported retaining age as a continuous linear term in the primary model.

Median regression and the influential-observation analysis yielded estimates consistent with the overall interpretation (Table 4). Fourteen observations had Cook’s distance greater than 4/n; after their exclusion, 177 participants remained. The historical-threshold logistic model reproduced the original adjusted estimate for AIDS at diagnosis (adjusted odds ratio, 1.77; 95% CI, 0.94 - 3.32; P = 0.075), which was treated only as a sensitivity analysis.

Table 4. Median-regression and influential-observation sensitivity analyses.

The influential-observation model excluded 14 observations with Cook’s distance > 4/n and used HC3 robust standard errors. Median regression estimates are reported at the 50^th percentile; the age and schooling coefficients were numerically close to zero and round to 0.000 at three decimals. CI indicates confidence interval.