Shifting neuropsychiatric diagnoses across the antiretroviral therapy era at an urban Ugandan HIV centre: a descriptive analysis of routinely collected electronic health records, 2017 to April 2026

Abstract

Background

As antiretroviral therapy (ART) coverage has matured in sub-Saharan Africa, the clinical profile of people living with HIV has shifted from acute opportunistic disease toward chronic neuropsychiatric morbidity. Few high-volume African HIV programmes have systematically described the full neuropsychiatric diagnostic architecture captured in their routine electronic records. We characterised the neuropsychiatric diagnoses recorded at Mildmay Uganda, an urban HIV centre serving approximately 14,500 active patients.

Methods

We conducted a descriptive analysis of routinely collected, de-identified electronic health-record data extracted from the ClinicMaster electronic medical record on 25 April 2026. Records with at least one ICD-10-coded neuropsychiatric diagnosis between 31 October 2017 and 23 April 2026 were grouped into nine clinically defined diagnostic categories using a pre-specified code list (Supplementary Table S1). Analyses were conducted at two levels: a record level (all 1,137 records) and a patient-index level (one earliest record per patient, n = 581). The design was cross-sectional and descriptive; no longitudinal follow-up, incidence or time-to-event analyses were undertaken. Proportions are reported with 95% Wilson confidence intervals where patients were the independent unit. Missing data were quantified rather than imputed.

Results

The dataset comprised 1,137 records from 581 patients (median 1 record per patient, IQR 1 to 2, maximum 13). Patients were 60.6% female (352/581) with a median age of 44 years (IQR 35 to 52). At the record level, four groups accounted for 90.6% of records: major depressive disorder (302; 26.6%), opportunistic CNS infection (293; 25.8%), HIV-attributed and other organic CNS disorders (256; 22.5%) and bipolar spectrum disorder (179; 15.7%). At the patient-index level, opportunistic CNS infection was the most common index presentation (206; 35.5%, 95% CI 31.7 to 39.4). A clinically coherent ART-duration gradient separated opportunistic CNS infection (median 0.2 years on ART at diagnosis) from chronic neuropsychiatric diagnoses recorded after several years of treatment (Kruskal-Wallis p < 0.001). Among the 406 index patients with viral-load data (69.9%), 82.5% (95% CI 78.5 to 85.9) were suppressed below 1,000 copies/mL. Record-level viral-load completeness ranged from 51.0% (2020) to 87.0% (2019) across the observation period, with an overall availability of 76.5% (870/1,137). CD4 count at ART initiation was unavailable for all records.

Conclusions

These routine data document a measurable treatment-era shift from acute CNS opportunistic disease toward chronic neuropsychiatric diagnoses among people living longer on ART. The findings are descriptive and hypothesis-generating, constrained by reliance on clinician-assigned ICD-10 codes. With standardised neuropsychiatric assessment, this routine-data platform can support prospective brain-health research and mental-health service integration in urban Ugandan HIV care.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s12981-026-00925-x.

Background

Mildmay Uganda is a Christian faith-based, non-governmental HIV care and research centre in Lweza, Wakiso District, approximately 12 km from Kampala. It provides comprehensive HIV care, including antiretroviral therapy (ART), mental-health services, neurological assessment and chronic-disease management, to approximately 14,500 active patients, most of whom are low-income urban and peri-urban adults from the wider Kampala metropolitan area. Routine clinical data are captured in the ClinicMaster electronic medical record (EMR), which stores ICD-10-coded diagnoses, ART prescriptions, laboratory results and free-text clinical notes, making the centre one of the more data-rich HIV platforms in the country for analyses based on routine data.

As ART coverage has expanded across sub-Saharan Africa, the clinical profile of HIV care has changed. Opportunistic infections and advanced immunosuppression have become less frequent, while chronic neuropsychiatric morbidity, including HIV-associated neurocognitive disorder (HAND), depression, bipolar spectrum illness and dementia, has become more prominent among people living longer on treatment [1–4]. In Uganda, the burden of depressive symptoms among adults with HIV and the relationship between cognitive function and HIV outcomes have been documented [5, 6]. However, few high-volume HIV clinics have systematically characterised the full neuropsychiatric diagnostic architecture represented in their routine data or appraised that architecture for prospective research.

Routinely collected, ICD-10-coded data, when linked to ART duration, regimen class and viral load, can map the diagnostic spectrum of neuropsychiatric disorders within an HIV programme and identify where more rigorous prospective measurement is needed [7, 8]. This study had three descriptive objectives: first, to describe the neuropsychiatric diagnostic spectrum recorded at Mildmay Uganda between 2017 and April 2026; second, to examine how diagnostic groups relate to treatment-era markers, particularly ART duration at the time of diagnosis; and third, to appraise the completeness and quality of these routine data for future analytic work. Here, data-platform appraisal denotes a structured assessment of variable completeness, internal consistency and laboratory linkage.

Methods

Study design

This was a descriptive, cross-sectional analysis of routinely collected electronic health-record data. No patients were followed forward in time, and no incidence, person-time or time-to-event analyses were performed; the work is therefore not a cohort study. Two units of analysis are distinguished throughout: the record level (each clinical record, allowing repeated records from the same patient) and the patient-index level (one earliest, or index, record per patient). The study is reported in accordance with the STROBE statement for cross-sectional studies and the RECORD extension for studies using routinely collected health data [10, 11]; completed checklists are provided as Supplementary Material.

Setting and participants

We included all records of HIV-positive patients with at least one ICD-10-coded neuropsychiatric, psychiatric or neurological diagnosis recorded in ClinicMaster between 31 October 2017 and 23 April 2026, the date of the most recent qualifying record before extraction on 25 April 2026. No lower or upper age limit was applied; the dataset therefore includes a small number of adolescents alongside adults. Diagnoses were those entered by attending clinicians during routine care and retrieved from the structured ICD-10 diagnosis field of the EMR; they were not re-adjudicated against case notes, and no separate research diagnostic interview or neurocognitive battery was administered. Eligibility was defined at the record level; the analytic dataset and its derivation are shown in Fig. 1.

Fig. 1.

Derivation of the analytic dataset (STROBE/RECORD flow). Data were extracted from the ClinicMaster electronic medical record on 25 April 2026. The 1,137 records from 581 patients were analysed at two levels: a record level (all records) and a patient-index level (one earliest record per patient). Viral-load data were available for 406 of 581 index patients (69.9%) and 870 of 1,137 records (76.5%)

Diagnostic grouping and the HAND category

Individual ICD-10 codes and their clinical labels were grouped into nine pre-specified research categories: major depressive disorder; opportunistic CNS infection; HIV-attributed and other organic CNS disorders (abbreviated HAND/HIV-CNS); bipolar spectrum disorder; psychosis spectrum; dementia spectrum; anxiety and stress-related disorders; substance use disorder; and other neurological or medical conditions. The complete code-to-category mapping, with record counts for every code, is provided as Supplementary Table S1 to allow full reproducibility.

The HAND/HIV-CNS category requires careful interpretation. As Supplementary Table S1 shows, it is built predominantly on ICD-10 F06 codes for mental disorders due to a known physiological condition, dominated by mood disorder due to a physiological condition (F06.31 and F06.32 together account for 184 of 256 records), together with psychotic, anxiety and delirium codes (F06.0, F06.2, F06.4, F05) and a small number of amnestic and encephalopathy codes (F04, G93.40). These are clinician-coded secondary, or organic, mental disorders attributed to a physiological cause, frequently used within an HIV clinic to denote HIV-attributable CNS disease. They do not constitute Frascati-defined HAND [12], because no standardised neurocognitive testing or functional staging was performed. The category is therefore best read as a clinician-coded construct of HIV-attributed and organic CNS disorders rather than as validated neurocognitive impairment.

Variables and definitions

ART regimen class was categorised as dolutegravir-based (DTG), efavirenz-based (EFV), or other or protease-inhibitor-based. Viral suppression was defined using the WHO programmatic threshold of fewer than 1,000 copies/mL. A three-level classification additionally distinguished a viral load below 200 copies/mL, low-level viraemia (200 to 999 copies/mL) and an unsuppressed result of 1,000 copies/mL or more [9]. ART duration at diagnosis was calculated as the interval between ART start date and the date of the index neuropsychiatric record. WHO clinical stage was dichotomised as stage I or II versus stage III or IV. Because clinically important characteristics, including age, sex, ART duration, WHO stage and viral suppression, may all shape the recorded diagnostic mix, these variables are presented descriptively by diagnostic group as a simple interpretive framework rather than as predictors in any causal model.

Data sources, cleaning and quality assurance

A single structured extract was drawn from ClinicMaster on 25 April 2026. Records were de-duplicated on the unique visit identifier; all 1,137 records carried distinct identifiers, so no duplicate records were removed. Date fields were parsed to a common format and checked for out-of-range values; diagnosis year was derived from the record date. ICD-10 codes were mapped to research categories using a fixed lookup table (Supplementary Table S1), and the mapping was checked manually against clinical labels. Continuous clinical variables were range-checked. Viral-load results entered as text (for example, target not detected) were harmonised with numeric results before classification. No values were imputed.

Missing data

The extent of missing data was quantified for all major variables overall, by calendar year and by diagnostic group (Supplementary Tables S2 to S4). At the record level, sex and age were complete; ART duration was missing for 0.4% of records, WHO stage for 11.6% and ART regimen class for 15.5%. Viral load was missing for 23.5% of records overall (267/1,137), with year-to-year variation ranging from 13.0% (2019) to 49.0% (2020). CD4 count at ART initiation was not retrievable for any record (100% missing) and is therefore not reported. Viral-load completeness was treated as a substantive data-quality finding rather than a statistical nuisance.

Statistical analysis

Analyses were descriptive. Categorical variables are summarised as counts and percentages, and continuous variables as medians with interquartile ranges (IQR). For proportions estimated at the patient-index level, where patients are the independent unit, 95% Wilson confidence intervals are reported. Confidence intervals are not attached to record-level proportions, because repeated records from the same patient breach the independence assumption; these are presented as descriptive frequencies. Patients could be coded into more than one diagnostic group across different visits; the extent of this overlap is quantified (Supplementary Table S4), and the patient-index analysis uses the earliest record only, so that each patient contributes once and to a single group. As a single pre-specified comparison, ART duration at diagnosis was compared across diagnostic groups using the Kruskal-Wallis test, reported as a descriptive test of difference and not interpreted causally. Analyses were conducted in R version 4.4.2.

Ethics and data governance

Ethical approval was granted by the Mildmay Uganda Research Ethics Committee (MUREC-2025-808; valid 29 April 2025 to 29 April 2026). Because the study used only de-identified data already collected during routine care, the committee granted a waiver of individual informed consent. Data were extracted by authorised health-information staff, de-identified at source by removal of names, clinic numbers and contact details, and analysed on password-protected institutional systems under the centre’s data-governance policy. No patient-level identifiers appear in this manuscript.

Results

Analytic dataset and the record versus patient distinction

The analytic dataset comprised 1,137 clinical records from 581 unique patients (Fig. 1). Most patients contributed a single record (median 1, IQR 1 to 2; maximum 13), and 258 patients (44.4%) had more than one record. Because the number of records per patient differed systematically by diagnosis, the diagnostic ranking differs between the two analysis levels, and both are reported. Recurrent conditions such as mood disorders generated repeated entries, whereas acute opportunistic CNS infections, often single severe episodes, generated fewer. This is why major depressive disorder leads the record-level count while opportunistic CNS infection leads the patient-index count.

Across the 581 patients, 352 (60.6%) were female and 229 (39.4%) male; the median age at the index record was 44 years (IQR 35 to 52; mean 43.6, SD 13.5).

Diagnostic spectrum

At the record level, four groups accounted for 90.6% of all records: major depressive disorder (302; 26.6%), opportunistic CNS infection (293; 25.8%), HIV-attributed and other organic CNS disorders (256; 22.5%) and bipolar spectrum disorder (179; 15.7%). Smaller groups comprised psychosis spectrum (45; 4.0%), dementia spectrum (31; 2.7%), anxiety and stress-related disorders (25; 2.2%), substance use disorder (4; 0.4%) and other neurological or medical conditions (2; 0.2%) (Fig. 2). The corresponding patient-index frequencies, with 95% confidence intervals, are shown in Table 1; opportunistic CNS infection was the most common index diagnosis (206; 35.5%, 95% CI 31.7 to 39.4), followed by HIV-attributed and organic CNS disorders (148; 25.5%, 22.1 to 29.2) and major depressive disorder (138; 23.8%, 20.5 to 27.4). Of the 581 patients, 150 (25.8%) were coded into more than one diagnostic group over time (Supplementary Table S5) (Fig. 3).

Fig. 2.

Record-level diagnostic spectrum based on ICD-10-derived research groups (n = 1,137 records). Patients with recurrent conditions contribute more than one record; patient-index frequencies, which differ in ranking, are given in Table 1

Table 1.

Patient-index demographic profile and diagnostic spectrum (one earliest record per patient; n = 581)

Diagnostic group (index)	Patients, n	% of patients (95% CI)	Female, n (%)	Age, years, median (IQR)
Opportunistic CNS infection	206	35.5 (31.7–39.4)	94 (45.6)	42 (35–49)
HIV-attributed/organic CNS (HAND/HIV-CNS)	148	25.5 (22.1–29.2)	111 (75.0)	41 (33–51)
Major depressive disorder	138	23.8 (20.5–27.4)	98 (71.0)	47 (35–53)
Bipolar spectrum disorder	40	6.9 (5.1–9.2)	27 (67.5)	48 (42–57)
Anxiety / stress-related	17	2.9 (1.8–4.6)	9 (52.9)	33 (19–47)
Dementia spectrum	15	2.6 (1.6–4.2)	4 (26.7)	63 (57–65)
Psychosis spectrum	12	2.1 (1.2–3.6)	5 (41.7)	41 (35–57)
Substance use disorder	4	0.7 (0.3–1.8)	3 (75.0)	41 (31–51)
Other neurological / medical	1	0.2 (0.0–1.0)	1 (100.0)	66
Total	581	100	352 (60.6)	44 (35–52)

CI confidence interval (Wilson), IQR interquartile range. Percentages of patients use the full denominator of 581. Female percentage is within each diagnostic group

Fig. 3.

Annual proportional composition of index diagnostic groups, 2017 to April 2026 (patient-index analysis; one earliest record per patient). Proportions in years with small denominators are unstable and should not be read as incidence

ART-duration gradient and regimen composition

A clinically coherent treatment-era gradient was evident across diagnostic groups (Table 2; Fig. 4). Opportunistic CNS infections were recorded close to ART initiation (median ART duration 0.2 years, IQR 0 to 2.2), consistent with the advanced immunosuppression that typically underlies CNS opportunistic disease. HIV-attributed and organic CNS disorders occupied an intermediate position (median 5.0 years, IQR 0.8 to 9.7) (Figs. 5 and 6). Mood, anxiety and dementia diagnoses were recorded after longer treatment exposure: major depressive disorder at a median of 7.5 years (IQR 3.5 to 10.9), bipolar spectrum disorder at 6.5 years (3.0 to 10.7), anxiety and stress-related disorders at 6.4 years (0.9 to 9.1), and dementia spectrum at 2.5 years (0.1 to 9.0). The difference in ART duration across groups was unlikely to be due to chance (Kruskal-Wallis H = 125.56, p < 0.001), a result that describes rather than explains the gradient. DTG-based regimens predominated across groups among patients with regimen class documented (Fig. 7), consistent with Uganda’s national transition to DTG-based first-line therapy during 2018 and 2019; regimen distribution should be read in light of that calendar transition rather than as a property of any diagnostic group.

Table 2.

Treatment-era and clinical characteristics by patient-index diagnostic group

Diagnostic group	ART duration at diagnosis, years, median (IQR)	Viral load available, n (%)	WHO stage III/IV, n (% of documented)	DTG-based ART, % of documented
Opportunistic CNS infection	0.2 (0-2.2)	105 (51.0)	66 (38.8)	76.0
HIV-attributed/organic CNS	5.0 (0.8–9.7)	113 (76.4)	15 (11.8)	82.0
Major depressive disorder	7.5 (3.5–10.9)	119 (86.2)	18 (14.0)	85.7
Bipolar spectrum disorder	6.5 (3.0-10.7)	33 (82.5)	6 (17.6)	79.5
Anxiety / stress-related	6.4 (0.9–9.1)	14 (82.4)	2 (12.5)	50.0
Dementia spectrum	2.5 (0.1-9.0)	9 (60.0)	6 (46.2)	72.7
Psychosis spectrum	2.6 (0.1–11.2)	9 (75.0)	4 (33.3)	75.0
Substance use disorder	4.0 (2.8–5.4)	3 (75.0)	1 (25.0)	100.0

ART antiretroviral therapy, DTG dolutegravir, WHO World Health Organization. ART duration was available for 99.5% of index patients. Kruskal-Wallis test for ART duration across groups: H = 125.56, p < 0.001. Other neurological/medical (n = 1) omitted

Fig. 4.

ART duration at the index neuropsychiatric diagnosis, by diagnostic group (patient-index analysis; other neurological/medical [n = 1] omitted). Boxes show median and interquartile range; individual observations are overlaid. Kruskal-Wallis H = 125.56, p < 0.001

Fig. 5.

Viral-load completeness by calendar year (record-level audit, n = 1,137). Completeness ranged from 51.0% in 2020 to 87.0% in 2019, with an overall availability of 76.5%. The 2020 dip likely reflects COVID-19-related service disruptions; partial years (2017, 2026) have smaller denominators

Fig. 6.

Virological suppression by index diagnostic group, restricted to patients with viral-load data (n = 406 of 581). Suppression uses the WHO threshold of fewer than 1,000 copies/mL. Wide confidence intervals for small groups (psychosis, dementia) indicate that differences are hypothesis-generating

Fig. 7.

ART regimen composition by index diagnostic group among patients with regimen class documented. Dolutegravir-based regimens predominated across groups, consistent with Uganda’s national guideline transition during 2018 to 2019. Regimen distribution should be read in light of the calendar period of diagnosis

Virological suppression and laboratory completeness

Viral-load data were available for 406 of 581 index patients (69.9%, 95% CI 66.0 to 73.5). Among these patients, 335 (82.5%, 95% CI 78.5 to 85.9) were suppressed below 1,000 copies/mL. Suppression among patients with data ranged from 76.2% in opportunistic CNS infection (95% CI 67.2 to 83.3) to 92.9% in anxiety and stress-related disorders (68.5 to 98.7) (Table 3; Fig. 6). The psychosis spectrum group had the lowest observed suppression (5 of 9 patients with data; 55.6%), but the confidence interval was very wide (26.7 to 81.1), and the group contained only 12 patients in total; this difference is therefore hypothesis-generating rather than evidence of a programme-level suppression gap. The 100% suppression observed in the dementia (9 patients) and substance use (3 patients) groups is similarly imprecise.

Table 3.

Virological suppression by diagnostic group, restricted to patients with viral-load data (n = 406)

Diagnostic group	Patients, n	VL available, n (%)	Suppressed < 1,000 c/mL, n (% of available, 95% CI)	VL < 200 c/mL, n (%)
Major depressive disorder	138	119 (86.2)	101 (84.9; 77.4–90.2)	---
Opportunistic CNS infection	206	105 (51.0)	80 (76.2; 67.2–83.3)	---
HIV-attributed/organic CNS	148	113 (76.4)	94 (83.2; 75.2–89.0)	---
Bipolar spectrum disorder	40	33 (82.5)	29 (87.9; 72.7–95.2)	---
Psychosis spectrum	12	9 (75.0)	5 (55.6; 26.7–81.1)	---
Dementia spectrum	15	9 (60.0)	9 (100.0; 70.1–100)	---
Anxiety / stress-related	17	14 (82.4)	13 (92.9; 68.5–98.7)	---
Overall	581	406 (69.9)	335 (82.5; 78.5–85.9)	---

VL viral load, c/mL copies/mL, CI Wilson confidence interval. Suppression uses the WHO threshold (< 1,000 c/mL). Wide intervals for small groups (notably psychosis, dementia and substance use) indicate that group-level differences are imprecise and hypothesis-generating. Substance use disorder (3 with VL) and other neurological/medical (1 with VL) omitted owing to very small denominators

At the record level, viral-load completeness was 76.5% overall (870/1,137) and ranged from 51.0% in 2020 to 87.0% in 2019 (Fig. 5). The lower completeness in 2020 likely reflects COVID-19-related service disruptions. CD4 count at ART initiation was unavailable for all records and is therefore not reported as a clinical variable.

Discussion

This descriptive analysis characterises the neuropsychiatric diagnoses recorded for 581 people with HIV at an urban Ugandan treatment Centre over an eight-and-a-half-year period. Its contribution lies in breadth: affective disorders, CNS opportunistic infections, HIV-attributed and organic CNS syndromes, bipolar illness, psychosis, anxiety and dementia diagnoses all appear within a single high-volume HIV platform, allowing a transparent description of what routine data at such a Centre can and cannot currently support.

The most clinically coherent observation is the ART-duration gradient. Opportunistic CNS infections were recorded near ART initiation, whereas chronic neuropsychiatric diagnoses appeared after several years of treatment. This pattern is consistent with advanced immunosuppression driving early CNS disease, and with immune reconstitution, neuroinflammation and ageing contributing to later morbidity [1–4, 12]. The association is cross-sectional and observed in routine data; it does not establish progression within individuals but describes the treatment-era stage at which different diagnoses are coded. Read this way, it converts a familiar clinical impression into a measurable description and helps identify which patient groups would be most informative for different prospective study designs.

The predominance of depression and HIV-attributed CNS diagnoses in this setting is consistent with reports from other sub-Saharan African HIV programmes, in which depressive symptoms and cognitive complaints are increasingly recognised as the ART era matures and survival improves [1–6]. Earlier Ugandan work documented a substantial burden of depressive symptoms and cognitive difficulty among adults with HIV initiating or established on treatment [5, 6]. Our findings extend that literature by describing, within a single routine-care platform, how these conditions sit alongside the declining but still present burden of acute CNS opportunistic disease. The contrast between our clinician-coded diagnoses and the standardised instruments used in dedicated neuro-HIV cohorts is itself informative, and motivates the case for embedding validated assessment within routine care.

The lower observed suppression in the psychosis group is best regarded as hypothesis-generating. With only nine patients contributing viral-load data, the estimate is imprecise, and any true difference could reflect the recognised relationship between severe mental illness, adherence and virological outcomes rather than programme failure [13]. If confirmed in larger samples, it would identify a group warranting targeted adherence support.

Viral-load completeness varied across the observation period, with a notable dip in 2020 (51.0%) likely attributable to COVID-19-related service disruptions, and somewhat lower completeness in partial-year periods (2017 and 2026). Year-to-year variation in laboratory linkage is a general feature of routine-data platforms and underscores the importance of ongoing data-quality monitoring.

Potential sources of bias

Several biases inherent to routinely collected diagnostic data require consideration. Referral and ascertainment bias are likely, because patients reaching a specialist mental-health or neurology service are not representative of all people with HIV, and only conditions that prompted a coded diagnosis appear in the data. Clinician coding behaviour introduces further variability: the choice between a primary depressive code and an organic mood-disorder code is clinician-dependent, and the validity of the HAND/HIV-CNS grouping is correspondingly limited. Temporal changes in coding practice and in the EMR itself mean that apparent year-on-year shifts partly reflect documentation rather than disease. Survivorship bias may also operate, because patients who died or disengaged are under-represented among those accruing later diagnoses. Finally, missing data are unlikely to be completely at random: viral-load availability differed by diagnostic group and by calendar year. Each of these considerations supports a descriptive reading of the findings.

Limitations

Further limitations should be noted. The analysis is single-centre and urban, which restricts generalisability to similar primary-care or hospital-based HIV settings. Diagnoses rest on clinician-assigned ICD-10 codes rather than structured psychiatric interviews or neuropsychological testing, so the HAND/HIV-CNS category in particular should not be equated with validated neurocognitive disorder. CD4 count at ART initiation was unavailable for all records, limiting characterisation of baseline immune status. Small group sizes for psychosis, dementia, substance use and other conditions yield imprecise estimates, as the confidence intervals make explicit. The cross-sectional design precludes any inference about incidence, progression or causation.

Implications and future directions

Within these constraints, the analysis points to three priority research questions that this platform is well placed to address. First, what is the true prevalence and severity of neurocognitive impairment among virologically suppressed adults on long-term ART at this centre, measured with a validated, locally appropriate instrument rather than routine codes? Second, which clinical and virological factors, including ART duration, regimen class and suppression history, are associated with the recorded neuropsychiatric diagnostic mix across treatment eras? Third, does embedding a structured brain-health screening and referral pathway into high-volume HIV care improve detection and management of neuropsychiatric morbidity? Pursuing these questions requires adopting a standardised coding and assessment standard for neuropsychiatric diagnoses and maintaining routine data-quality monitoring.

Conclusions

Routine electronic records at Mildmay Uganda document, in measurable terms, a treatment-era shift from acute CNS opportunistic disease toward chronic neuropsychiatric diagnoses among people living longer on ART. The findings are descriptive and hypothesis-generating, bounded by reliance on clinician-assigned ICD-10 codes. With standardised neuropsychiatric assessment, this routine-data platform can support prospective brain-health research, mental-health service integration and health-information-system strengthening in urban Ugandan HIV care.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

ART

Antiretroviral therapy

CI

Confidence interval

CNS

Central nervous system

DTG

Dolutegravir

EFV

Efavirenz

EMR

Electronic medical record

HAND

HIV-associated neurocognitive disorder

ICD-10

International Classification of Diseases:10th revision

IQR

Interquartile range

LIMS

Laboratory information management system

VL

Viral load

WHO

World Health Organization

Author contributions

JM conceived the study, developed the analysis plan, led data analysis and interpretation, and drafted the manuscript. CS contributed to study conceptualisation, interpretation of clinical and programme implications, and critical manuscript review. JN contributed to data interpretation, contextualisation of findings within routine clinical care, and critical manuscript review. RM contributed to research governance, institutional interpretation, and critical manuscript review. YK contributed to institutional oversight, interpretation of health systems implications, and critical manuscript review. All authors reviewed and approved the final manuscript.

Funding

No specific external funding was received. The analysis was conducted as part of institutional research and development work at Mildmay Research Centre Uganda.

Data availability

The dataset analysed for this manuscript consists of de-identified routine clinical data from Mildmay Uganda. The data are not publicly available because they contain sensitive HIV, neuropsychiatric and clinical service information governed by institutional data protection and research ethics requirements. Reasonable requests for access to aggregated outputs or further information may be directed to the corresponding author and will be considered by Mildmay Research Centre Uganda subject to institutional approval, ethical clearance and applicable data-sharing agreements.

Declarations

Ethics approval and consent to participate

Ethical approval was obtained from the Mildmay Uganda Research Ethics Committee (MUREC-2025-808; approval period 29 April 2025 to 29 April 2026). The committee granted a waiver of individual informed consent because the study used only de-identified routine clinical data.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.