Selection criteria and study outcomes

A detailed overview of inclusion and exclusion criteria is provided in Supplementary material, Table S2. Briefly, we included blinded and open‑label randomized controlled trials (RCTs) that compared tenofovir alafenamide– or tenofovir disoproxil fumarate–based ART or PrEP with any pharmacological control, including placebo. As we were interested in the consequences of long‑term medication, trials with a minimum duration of 2 months were eligible. To avoid preselection bias due to tolerability and to reduce heterogeneity, only trials involving therapy‑naive adults were included. Our prespecified outcomes of interest were types of NPAEs and the number of affected patients.

Data extraction and quality assessment

The identified records were screened and assessed for eligibility by 2 independent reviewers (MG, MSS). Extracted data included publication details, study design and duration, sample characteristics, treatment characteristics, and types and number of NPAEs. In the case of multiple reports on a single trial, data were collected from the record with the most complete AE report. Data from different dosing arms within the same study were pooled. A previously published checklist¹⁷ was used to rate the quality of AE measurement and reporting. Two reviewers (MG, MSS) independently performed a quality assessment for all included trials using the revised Cochrane risk‑of‑bias tool for randomized trials (RoB 2).¹⁸ Any disagreements were resolved through a discussion with a third reviewer (TGR).

Qualitative analysis

The aim of the qualitative analysis was to identify common NPAEs occurring in patients on tenofovir‑based regimens. Types of NPAEs were defined a priori through a discussion between neurologists, psychiatrists, and clinical pharmacologists. The prespecified list of NPAEs was open to extension based on additional symptoms reported in trials and deemed relevant by our interdisciplinary team (Supplementary material, Table S3). Symptoms were clustered if they 1) were synonymous or could be grouped under a single umbrella term, and 2) did not occur within the same study arm. Incidences of NPAEs were aggregated within and across treatment regimens, and NPAEs with at least 50 occurrences in HIV trials were regarded as common. In addition to evaluating the frequency of common NPAEs, data on the relationship between treatment duration and NPAE emergence, as well as withdrawals due to NPAEs, were sought.

Statistical analysis

The quantitative analysis was structured around 2 key comparisons. The first one concerned tenofovir’s effect: meta‑analyses were conducted to compare NPAE outcomes of tenofovir use (alone or with emtricitabine) against a placebo. The inclusion of emtricitabine reflected the clinical practice that tenofovir was rarely used as monotherapy. The second comparison was focused on identifying differences between tenofovir disoproxil fumarate and tenofovir alafenamide: we analyzed studies that included both a tenofovir alafenamide–based intervention and a tenofovir disoproxil fumarate–based comparator to assess potential differences in the risk of NPAEs between the tenofovir prodrugs.

We included both open‑label and blinded studies across indications (ART, PrEP). Meta‑analyses were performed if 3 or more studies reported NPAEs using the same term with at least 10 cumulative events in every treatment group. Analyses were conducted using Review Manager 5.4 (Cochrane Collaboration, London, United Kingdom).¹⁹ We calculated odds ratios (ORs) with 95% CIs using a random‑effects model due to expected heterogeneity in the study population and design. We used a P value of <⁠0.05 to determine statistical significance. Due to the exploratory nature of our study, P values were not adjusted for multiple comparisons. Heterogeneity in the effect sizes was assessed using the I² statistic.

We planned 3 sensitivity analyses for both comparisons: 1) exclusion of studies with the highest weight, 2) exclusion of studies with a high risk of bias, and 3) leave‑one‑out analyses. In addition, for the comparison of tenofovir (with / without emtricitabine) vs placebo, an additional sensitivity analysis was performed: 4) restriction to studies utilizing tenofovir as monotherapy. Three sensitivity analyses were intended to be performed for the comparison between tenofovir disoproxil fumarate and tenofovir alafenamide: 5) exclusion of open‑label studies, 6) restriction to studies that administered the same concomitant medication in both tenofovir arms, and 7) restriction to HIV studies.

Ethics

Ethical approval and patient consent were not required for this systematic review, as all data were extracted from previously conducted studies and could not be attributed to individual participants.

Search results

Following duplicate exclusion, a total of 8058 records were identified in our search (Supplementary material, Figure S1). After screening titles and abstracts, we assessed 1186 full‑text papers for eligibility. Ultimately, 69 RCTs (62 on ART and 7 on PrEP) were included in the qualitative analysis. Four trials were unpublished, with data reported only on trial registries. Additional data were obtained from authors in 6 cases.

Characteristics of eligible studies on antiretroviral therapy in the qualitative synthesis

The 62 included ART studies involved 23 016 patients on tenofovir‑based therapy: tenofovir alafenamide in 11 studies with 2997 patients, and tenofovir disoproxil fumarate in 57 studies with 20 219 patients. Of these, 34 studies were double‑blind and 28 were open‑label. One study involved HIV and hepatitis B–coinfected patients. The reported mean age ranged from 26 to 41.6 years. Women accounted for only 19% of the study population; however, 2 studies exclusively enrolled women. The main characteristics of all included ART trials are summarized in Supplementary material, Table S4.

Across all ART trials, 27 individual therapeutic agents were employed in 29 tenofovir‑based combinations. Tenofovir alafenamide was only combined with emtricitabine and protease inhibitors, integrase strand transfer inhibitors, or, in 1 case, a capsid inhibitor. Tenofovir disoproxil fumarate was combined with either other nucleoside reverse transcriptase inhibitors or non‑nucleoside reverse transcriptase inhibitors.

Characteristics of eligible studies on pre‑exposure prophylaxis in the qualitative synthesis

We included 7 trials focused on PrEP, investigating tenofovir disoproxil fumarate either as monotherapy or with emtricitabine in 6324 patients. One study also analyzed tenofovir alafenamide with emtricitabine in 2694 patients. Women accounted for 43% of the study population across all PrEP studies. The mean age was reported in only 2 studies (23.5 and 36 years, respectively). A detailed overview of the PrEP trials is presented in Supplementary material, Table S5.

Quality assessment and bias rating

Of the ART studies, only 9 detailed the AE assessment methods. Although data from trial registries or original investigators were used to augment the AE reports, only a quarter of the ART trials had complete AE data without preselection for frequency. Bias risk in the blinded ART trials was generally low, except for studies involving efavirenz, which entailed a higher risk due to potential unblinding from characteristic NPAEs (Supplementary material, Figure S2). Open‑label studies were generally assessed to have a high risk of bias (Supplementary material, Figure S3).

Among the PrEP studies, only 1 provided information on the AE collection method. Complete AE reporting was limited to 2 studies. Four of the 7 studies were judged to have a moderate or high risk of bias, primarily attributed to open‑label designs or inadequate details on the randomization process (Supplementary material, Figures S4 and S5).

Qualitative analysis

More than 200 NPAEs were reported across all tenofovir‑based treatment arms, but only 13 occurred in more than 50 patients each (Supplementary material, Table S6). The most frequent neurologic events were headache (3071 affected of 21 608 exposed patients), dizziness (2242 of 17 633), somnolence (305 of 7597), vertigo (110 of 3776), paresthesia (92 of 3976), and peripheral neuropathy (80 of 3437). The most frequent psychiatric events were insomnia (1476 of 16 399), depression (1229 of 16 085), abnormal dreams (847 of 8577), anxiety (679 of 11 105), decreased appetite (249 of 7214), sleep disorders (143 of 4625), and nightmares (87 of 3351). Notably, efavirenz‑containing treatment groups accounted for a large number of cases across most of these events.

The spectrum of reported NPAEs during PrEP was narrower, and relative frequencies for most observed events were numerically lower. Dizziness, headache, insomnia, and depression were the only NPAEs with aggregated relative frequencies of at least 2% across all reviewed studies (Supplementary material, Table S7).

The relationship between treatment duration and NPAE frequency was rarely explored in the included studies. One PrEP study comparing tenofovir disoproxil fumarate with placebo reported monthly dizziness rates over 24 months, with the highest incidence in the first month of therapy.²⁰ Furthermore, some ART studies provided data for different observation periods. For example, a study comparing 2 tenofovir disoproxil fumarate regimens found most new cases of headache and depression emerging at weeks 1–48, though new cases continued to appear during follow‑up.^21,22

Twenty‑five studies providing complete reports of AEs leading to withdrawal were used to calculate the rate of withdrawals due to NPAEs. In the ART studies, dizziness and depression were the most common neurologic and psychiatric AEs leading to withdrawal, with 11 and 13 cases, respectively, out of 675 and 415 patients experiencing these NPAEs. In the PrEP studies, only isolated cases of withdrawal were reported (Supplementary material, Table S8).

Meta‑analyses of tenofovir vs placebo

Five double‑blind RCTs using tenofovir disoproxil fumarate for PrEP met the criteria for inclusion in the meta‑analysis. Of these, 3 used tenofovir disoproxil fumarate alone, whereas emtricitabine was added in 2 trials. Meta‑analyses were possible for 4 neuropsychiatric symptoms (Figure 1). Analyses for headache (OR, 0.96; 95% CI, 0.85–1.09; P = 0.56), anxiety (OR, 0.73; 95% CI, 0.41–1.31; P = 0.3), and insomnia (OR, 0.93; 95% CI, 0.75–1.16; P = 0.1) yielded no differences between tenofovir disoproxil fumarate–based PrEP and placebo. However, a higher risk for dizziness (OR, 1.32; 95% CI, 1.09–1.59; P = 0.004) was observed in the tenofovir disoproxil fumarate group. Heterogeneity was only detected for anxiety, where it was low (I² = 11%).²³ These results were mostly supported by our sensitivity analyses (Supplementary material, Table S9). Notably, when the analyses were limited to studies using tenofovir disoproxil fumarate as monotherapy, the observed effects on dizziness diminished to a nonsignificant trend (OR, 1.29; 95% CI, 0.98–1.68; P = 0.06).

Meta‑analyses of tenofovir disoproxil fumarate– vs tenofovir alafenamide–based regimens

Six double‑blind studies on ART and 1 on PrEP were eligible for a comparison of the tenofovir prodrugs. Meta‑analyses were possible for dizziness (OR, 0.87; 95% CI, 0.6–1.24; P = 0.43), headache (OR, 1.1; 95% CI, 0.95–1.27; P = 0.19), anxiety (OR, 1; 95% CI, 0.671.5; P = 0.99), depression (OR, 1.25; 95% CI, 0.91–1.72; P = 0.18), and insomnia (OR, 1.27; 95% CI, 0.96–1.68; P = 0.1) (Figure 2). None of the comparisons yielded significant differences between tenofovir alafenamide and tenofovir disoproxil fumarate. Heterogeneity was absent for all symptoms (I² = 0%) except for anxiety, where it was low (I² = 21%).²³ Sensitivity analyses were mostly unremarkable, except for that involving headache (Supplementary material, Table S10). After exclusion of the single PrEP study, a higher headache risk associated with tenofovir alafenamide–based therapies was observed (OR, 1.24; 95% CI, 1.01–1.52; P = 0.04; I² = 0%). This study was excluded from 3 sensitivity analyses: exclusion of the study with the highest weight, leave‑one‑out, and restriction to HIV therapy studies.

Limitations

Importantly, to our best knowledge, specific NPAEs other than headache and insomnia have not been investigated in previous meta‑analyses of tenofovir alafenamide– vs tenofovir disoproxil fumarate–based regimens. Considering the increasing significance of tenofovir alafenamide in ART, as well as its role in PrEP, our finding of noninferiority to tenofovir disoproxil fumarate regarding most NPAEs represents an important contribution to an understanding of its risk profile.^15,35

Several limitations must be considered when interpreting the results of our study. First, to reduce possible investigator bias, we extracted the number of all AEs vs only drug‑related AEs, meaning that some observed events may not be causally related to the treatment. This is especially true in the ART studies, where the high prevalence of neuropsychiatric conditions in PLWH²⁷ may have led to an overestimation of the effect of tenofovir‑based therapies on neuropsychiatric health and contributed to the higher numbers and frequencies of NPAEs, as compared with the PrEP trials.

Second, due to incomplete AE reporting in most included studies, the total number of reported cases may not represent the actual numbers for some NPAEs. Frequency thresholds for AE reporting present in most studies omitted rarer events, and many records were excluded due to missing, preselected, or grouped AE data. We tried to correct this by contacting the investigators but obtained additional data for only 6 trials.

Third, our study employed stringent inclusion criteria, specifically focusing on treatment‑naive patients. This methodological choice serves as both a strength and a limitation of our meta‑analyses. In contrast to a previous meta‑analysis that compared the general safety of tenofovir disoproxil fumarate– vs tenofovir alafenamide–based therapies, which incorporated studies including treatment‑experienced patients,³⁶ our targeted approach on treatment‑naive patients was adopted to reduce potential biases associated with preselection for tolerability. This strategy aimed to improve homogeneity, while recognizing the trade‑off of potentially diminishing statistical power. Notably, this criterion also prompted our decision not to include studies focusing on the treatment of hepatitis B, which otherwise would have been valuable for both the comparison of tenofovir vs placebo and the comparison of the tenofovir prodrugs. Treatment‑naivety is rarely a prerequisite in hepatitis B trials, as most studies allow both treatment‑naive and treatment‑experienced patients.^37,38

Fourth, we pooled data on AEs reported up to different time points of observation and study duration for the meta‑analyses. This approach ensured inclusion of the most complete datasets available at the expense of introducing additional heterogeneity.

Finally, our decision to compare tenofovir alafenamide with tenofovir disoproxil fumarate stemmed from the hypothesis that higher intracellular tenofovir concentrations, potentially achieved with tenofovir alafenamide, might increase neurotoxicity.³⁹ To date, there has been no direct comparison of neuronal concentrations of tenofovir between these prodrugs. However, a single small study assessed tenofovir concentrations in cerebrospinal fluid (CSF) and plasma before and after switching from tenofovir disoproxil fumarate to tenofovir alafenamide in ART patients, using the CSF / plasma concentration ratio as a surrogate marker of brain penetration. The authors found that this ratio increased after the switch, suggesting enhanced brain penetration.⁴⁰ However, the small sample size and reliance on an indirect method to estimate brain penetration limit the generalizability of these findings. Without direct evidence comparing neuronal uptake, the question of increased neurotoxicity with tenofovir alafenamide remains open. Given its growing use, a thorough investigation into potential risks is warranted. Our analysis contributes to this exploration, emphasizing the need for cautious interpretation and further research to elucidate the neuropsychiatric safety profile of tenofovir alafenamide.

Conclusions

This study demonstrated that several types of NPAEs are common in tenofovir‑based ART. In the PrEP trials, where tenofovir was used alone or with emtricitabine, the spectrum and frequency of NPAEs were generally lower. The higher risk in the ART studies may reflect both the increased vulnerability of PLWH to NPAEs and the effects of other drugs included in ART regimens. Although NPAEs rarely led to therapy withdrawal in RCTs on tenofovir‑based therapy, such symptoms can impair adherence in real‑world settings,^3-5 underscoring the need to monitor these events. The European AIDS Clinical Society (EACS) guidelines recommend assessing depression and anxiety through leading questions,⁴¹ and we suggest expanding this list to include other common neuropsychiatric symptoms. The EACS guidelines also provide algorithms for managing neuropsychiatric symptoms. Importantly, if a causal link to ART is likely, switching regimens is warranted; otherwise, general guidelines (eg, for headache, depression) should be followed.⁴¹ Among NPAEs, dizziness may be the only symptom directly attributed to tenofovir. Tenofovir alafenamide is noninferior to its predecessor regarding NPAE rates, except for a potentially higher risk of headache in the ART patients, which warrants further investigation.