In this analysis using data from APHOD, neurocART was not significantly associated with a reduction in survival for HIV-positive patients, and this finding was consistently obtained across a range of sensitivity analyses. Similarly, a nonsignificant association was observed when the first incidence of ADI was incorporated as an endpoint, and no association was found between neurocART use compared with cART use and CD4 cell count. At least in APHOD, a potential benefit associated with neurocART use is not evident in overall population survival.
The use of neurocART has been shown to improve survival after diagnosis of HIV encephalopathy in perinatally infected children and adolescents [1
], but survival effects are less clear in general HIV-positive populations [21
]. Our analysis does not confirm the association of neurocART use and improved survival in a broader population of HIV-infected adults, with our findings being robust to changes in model assumptions. Further, the independent associations between other population and treatment characteristics and survival in our study were consistent with other findings [18
]: higher CD4 cell count was strongly associated with reduced mortality, while increased HIV viral load, increased age, certain modes of exposure (IDU and `other'), hepatitis coinfection, ADI and more extensive treatment history (higher regimen count) were associated with increased mortality. In the multivariate analysis, it was found that higher CD4 cell count, increased age, HBV coinfection, ADI and more extensive treatment history (higher regimen count) were strongly associated with survival after adjustment for other covariates. In terms of survival prediction, neurocART was not very important to the models in comparison with these covariates.
We did not directly examine NCI-associated mortality, although an important rationale for this study was the possible improvement in survival attributable to the beneficial effect of neurocART on mild, and possibly undiagnosed and unmeasured, NCI [1
]. Although previous studies have demonstrated a sizeable frequency of mild NCI in certain populations [8
], we do not have comprehensive data on the incidence of mild NCI-associated mortality in APHOD. To our knowledge, there is no strong existing evidence of survival attributable to the beneficial effects of neurocART on mild NCI. A recent paper by Smurzynski et al
] showed an adjusted association between increases in CPE score and neuropsychological test scores when accounting for an interaction with the number of ARVs per regimen. While Patel et al
. did not find a significant association between CNS penetration and the incidence of HIV encephalopathy, they did observe a significant survival benefit associated with CNS penetration in HIV encephalopathy cases [1
]. In contrast, while Garvey et al
. did not observe a significant adjusted association between CPE score and CNS opportunistic diseases, they noted that the lowest and highest CPE scores were associated with increased mortality [21
], but suggested that this was a consequence of clinical status affecting prescribing practice. Overall, our findings do not demonstrate the posited association between neurocART-reduced NCI and improved survival in APHOD.
Our findings, which describe prospective data for the period 1999–2009, can be contrasted with those of a recent study by Lanoy et al
], where all-cause mortality in neuroAIDS diagnoses was associated with CPE score for each of the periods 1992–1995 and 1996–1998 but not for 1999–2004. In that study, the authors attributed the lack of an associated effect in the period 1999–2004 to improved control of plasma viral load (which was not adjusted for in initial models) by cART regimens in general. In the same study, a secondary analysis for the period 1997–2004 showed no change in survival associated with CPE score after including plasma HIV RNA as a covariate. While our results reflect a lack of a differentiable survival effect of neurocART use in the later cART period for all HIV-positive patients, they also suggest that plasma viral load adds little extra descriptive power after the inclusion of CD4 cell count as a covariate in multivariate models when examining neurocART survival outcomes. Similarly, while Patel et al
. were unable to adjust for viral load in their primary analysis, sensitivity analyses suggested that measured CNS effects were not confounded by the omission of this covariate [1
]. In this regard, temporal changes in the measured CPE effect as observed by Lanoy et al
. are not necessarily reflected by measured HIV RNA suppression in these analyses, at least when also adjusting for CD4 cell counts. Nevertheless, broader changes in therapy, including general increases in cART CPE levels and potency, may reduce the effectiveness of CPE as a measure of neuroAIDS treatment, and wider changes in therapy should be considered in association with CPE measurements to describe the effectiveness of treatments of neuroAIDS.
Of note is the fact that in our study we used the 2010 CPE ranking approach, as presented by Letendre et al
]. While this approach has not been validated at the time of submission, we have found analysis results to be qualitatively similar to those obtained using the 2008 approach [16
] (data not shown). There are acknowledged weaknesses with the CPE scoring system, including scarce information on ARV CNS penetration and pharmacodynamics, including possible insensitivity to drug–drug interactions, the role of blood–brain barrier permeability in CNS drug penetration and the possible effects of ageing. However, the CPE scoring system represents a practical tool with which to assess CNS effectiveness of cART regimens and has been associated with strong measured improvement in overall survival in one study [1
]. As stated, a posited reason for this is that treatment of mild undiagnosed NCI with neurocART improves overall survival, although we were not able to evaluate this in our analysis. Furthermore, we were not able to evaluate the relationship between use of neurocART and cerebrospinal fluid HIV viral load results.
In APHOD, HAD and PML events are too rare to be used as statistical endpoints and detailed data on other neurological events are not collected; however, we looked at broader outcomes for neurocART use. The composite endpoint of `ADI or death' showed a weaker association, suggesting that neurocART use does not reduce the incidence of ADI compared with cART. Also of note is the finding that neurocART use was not strongly associated with changes in CD4 cell count compared with cART use. These findings do not demonstrate any additional benefit associated with neurocART use compared with non-neurocART use.
We also examined survival attributable to neurocART across different stages of treatment: for baseline neuro-cART, subsequent neurocART, and cumulative duration of neurocART. We observed a nonsignificant association between neurocART as the first cART and survival, consistent with the findings of Garvey et al
], where baseline CPE category was categorized as a four-level variable. In the same study, Garvey et al
. found that the lowest and highest categories of the latest CPE were associated with increased mortality in multivariate models; however, we did not find an equivalent association in APHOD. We also found that models using the latest neurocART showed a stronger, but still nonsignificant, association with survival than equivalent four-level CPE models. Finally, we observed no association between survival and duration of neurocART use when data were adjusted for other patient characteristics. Overall, we were unable to demonstrate a difference in survival associated with neurocART compared with non-neurocART.
There are several limitations to this study. Firstly, our study may have been underpowered to detect a significant association between CPE score and overall survival. Sample size calculations estimate that we would have needed over 1000 events to detect a significant improvement in survival of<15%. The likely low incidence of death associated with NCI further limits the power of analysis. In APHOD, the low incidence of HAD precluded it from being analysed directly, and limited data are collected on other NCI outcomes. Although APHOD comprises relatively large multisite cohorts with good follow-up, these results flag the need for more extensive data for examination of neuro-cART outcomes including associated mortality. In particular, examination of mild CNS events might increase the sensitivity of analyses to general neurocART outcomes including associated mortality, subject to available data and the constraints this places on the power of analyses. Although TAPHOD does not collect these data in any standardized fashion, we are not aware of any other cohorts that do so. In this regard, the routine screening for HIV-associated neurocognitive disorders in relevant cohorts should be considered. Similarly, although previous studies have identified clade-specific differences in HIV neurotoxicity [26
], our analysis did not specifically adjust for this. Differences in neurotoxicity by clade may potentially limit the general application of CPE as used in this analysis, and the inclusion of clade as a covariate to examine this should be considered in future analyses. Other limitations include the enrolment of patients in APHOD after the initiation of cART, and the enrolment of patients with mono/dual therapy experience prior to starting cART. To address these concerns, prior treatment experience was factored into analyses including prior treatment type, neurocART-first cART, regimen count and neurocART exposure. Of these covariates, only higher regimen counts (≥4 regimens) were found to contribute significantly to multivariate models.
In summary, our findings do not show a significant overall survival benefit associated with neurocART compared with cART in a population of HIV-positive adult patients (APHOD). In particular, the potential benefit associated with neurocART in terms of prevention of neurocognitive impairment did not translate into an improvement in overall survival in this population. These findings were limited by the likely low incidence of NCI-associated mortality. Further studies and more extensive data are needed to address these limitations.