Prior HIV cohort analyses have reported a disproportionately large statin-associated mortality benefit of > 50% [9–11] which resembles reports of ≥ 40% reduced mortality among statin users in other populations with altered immunity [27–29], inherently increased (cardiovascular) mortality risk [30–33], or old age (25% mortality reduction in men > 75 years) . Decreased mortality had never been observed in primary NS-LLT prevention trials but has recently been reported when icosapent-ethyl (fish oil component) or alirocumab (PSK-9 inhibitor) was added to statins in high-risk populations [35, 36].
The relationship between density of longitudinal LLT exposure and clinical effectiveness is incompletely understood. It could hinge on magnitude of cumulative exposure, consistency of exposure, and recency of use. To capture optimal exposures, “consistent use” in our multi-level exposure model required both > 91% adherence for ≥ 1 year and use within 30 days. To our knowledge, LLT effectiveness has not been analysed this way in high-risk populations. Multi-level time-updated drug exposure models have been tested , can address frailty bias , and are not subject to immortal time bias [14, 38]; both of which are known to lead to inflated treatment effects [14, 22].
The lack of a mortality benefit for remote LLT use argues against healthy user bias  and the lack of any benefit for consistent antihypertensive or aspirin use against healthy adherer bias  as explanations for the apparent mortality benefit of ongoing LLT use. The magnitude of the mortality benefit during consistent statin-free LLT use was unexpected and contrasted sharply with only moderately reduced mortality risk for inconsistent use – for which no reduced ASCVD risk was observed. Increased intra-individual (visit-to-visit) serum cholesterol variability has recently been identified as an important ASCVD and mortality risk factor [41, 42]. Albeit not yet biologically understood, this phenomenon could potentially offset beneficial LLT effects in patients with low adherence and may even play a role in randomized controlled trials of LLT.
For statins, the mortality difference between consistent and inconsistent use was much smaller. This may reflect their sustained immunomodulatory properties, as evidenced by reduced infection and cancer risk even for inconsistent, respectively remote users.
Our mortality model met consistency, positivity, and correctness of weight-generation criteria of marginal structural models . Similar reductions for overall ASCVD risk during consistent statin-free LLT and coronary risk during consistent statin-only LLT provide biologic plausibility for the reduced mortality risk. Yet after IPW and multi-level adjustment, consistent use of antihypertensives and aspirin remained associated with increased mortality and other adverse outcomes which indicates residual indication bias. This “stubborn” residual bias  was directed against patients taking cardiovascular preventive medications and would have affected statins similarly. As statins are arguably the most important preventive cardiovascular drug class, this residual bias may explain why they appeared less effective than statin-free LLT in reducing mortality. Yet, the lack of a cerebrovascular effect during consistent statin only LLT use is also noteworthy.
The current HAART era is characterized by high adherence to single tablet regimens, sustained virologic suppression, and durable immune restoration. We included patients only after achieving virologic suppression but continued to follow them regardless of virologic failure to avoid informative censoring. We further approximated contemporary conditions in contrasting subgroup analyses and observed comparable results. Consistent combination LLT use remained associated with significantly reduced mortality in all examined subgroups including patients with sustained virologic suppression and immune reconstitution and patients with low ASCVD risk. A notable exception were patients taking TDF containing HAART for whom the mortality impact of consistent LLT was attenuated. TDF (but not tenofovir alafenamide fumarate ) has well documented lipid lowering properties  and was the only ARV component independently associated with reduced mortality. Importantly, it is no longer used in most modern single-tablet HAART regimens.
There was no apparent association between absolute serum LDL levels and clinical outcomes within the same LLT exposure levels. But if the decreasing mortality risk from remote to consistent LLT exposures is interpreted as “dose-response relationship”, our study would fulfil most of the Bradford-Hill criteria  for causal inference between LLT use and mortality risk in PLWH. The REPRIEVE trial [17, 18] will provide the ultimate guidance on statin use in PLWH. But as unmeasured or uncontrolled confounding is unlikely a major explanation for our findings, broad use of lipid lowering therapy in HIV-infected US-veterans in past decades, including those without virologic suppression, could have saved thousands of lives.
The major strengths of our study are its comprehensiveness, its detailed drug exposure models, and its statistical approach. Others include cohort size and diversity, length of follow-up, and the reliance on uniform data collection on exposures and outcomes across the entire US-VA system. Limitations include an extreme male predominance, the lack of differentiation between different daily doses and the absence of cause of death. Its major limitation is the remote timeframe. But before the publication of the 2013 AHA/ACC Cholesterol guidelines , non-statin lipid lowering agents were commonly combined with or substituted for statins to target risk-specific cholesterol goals . This allowed our comparative analysis of different forms of LLT.