This paper is the first to provide estimates of long-term toxicity related to D4T up to six years after treatment initiation from LMICs. By six years on D4T, lipoatrophy was the predominant side-effect, requiring treatment substitution in seven out of ten patients. For each of the different D4T-toxicities, distinct risk factors were identified, indicating which patients might require more close toxicity-monitoring, should avoid D4T altogether or should be prioritized for phasing-out of D4T.
Our findings of older age and female sex as risk factors for lipoatrophy are consistent with other reports from LMIC 
. Regarding baseline CD4 cell count, conflicting data have been reported 
. Our data confirm the reported association of older age and neuropathy 
. Whereas most previous studies have reported advanced HIV stage as a risk factor for neuropathy, we observed an increased risk associated with low baseline haemoglobin. Possibly, this is merely a reflection of advanced disease at ART initiation 
. Alternatively, recent studies have observed the strong and independent prognostic information contained in both baseline and time-updated haemoglobin levels, even after adjustment for CD4 cell count values 
. In contrast with a South African study 
, we did not find an association of neuropathy with TB treatment. Possible reasons for this could include differences in analytical approach, management of toxicity or timing of ART initiation for patients on TB treatment.
With regards to SH/LA, the association with the use of tuberculosis treatment is somehow surprising and has not been reported yet. In one case-control study, efavirenz was identified as a risk factor for lactic acidosis 
. Whether this could be the mechanism behind our observed association with tuberculosis treatment, albeit not identified in multivariate analysis, remains to be determined. Alternatively, rapid weight gain after ART initiation, identified as a risk factor for lactic acidosis in a South African study, might be implicated for patients on tuberculosis treatment 
. Finally, it needs to be pointed out that TB treatment and the use of EFV are closely related variables. Although no clear problem of collinearity was detected during analysis, the problem cannot be entirely ruled out.
Cost has been a major reason for the ongoing use of D4T-containing ART in LMIC. Despite recent cost reductions, tenofovir-based regimens are still more than twice as expensive in terms of drug costs. In Cambodia, the low cost of D4T is a key argument for maintaining this drug within first line treatment regimens for the next years to come. Our data suggest that the cost-saving effect with D4T-use is limited in time, given the high long-term rates of D4T-replacement. Moreover, its ongoing use continues to expose patients to drug toxicity, with all its negative implications. Whereas our data reinforce the need to phase-out D4T to better tolerated regimens in LMIC, it is clear that this is a major operational undertaking that should be implemented in a phased and controlled manner. In this regard, our experience could be of interest for national programs willing to implement a gradual phasing-out of D4T. By combining patient education, close monitoring for D4T-toxicities, integrating the patient's perception and applying a low threshold for D4T-replacement, a gradual phasing-out of D4T can be expected. Additionally, the occurrence of toxicity could be significantly reduced by prioritizing those at highest risk of D4T toxicity, based on the risk factors identified. Importantly, patient support relative to a uniform and quick D4T replacement strategy might be enhanced with this more targeted approach. A recent study from Cambodia on systematic substitution of zidovudine for D4T highlighted that a fraction of patients preferred to remain on D4T-based HAART 
. Imposing a treatment change without patient approval might negatively affect on the adherence to the new regimen.
Previous studies have reported on D4T-toxicity up to three years after ART initiation 
. Although incidence of neuropathy in our study was lower, our three-year estimates on lipoatrophy are around twice those recently reported in an African cohort 
. This could in part relate to the strict criteria for D4T-substitution in this program, relying on a scoring-system of clinical severity of changes in different body sites. In contrast, we also considered the patient's perception and preference. Indeed, recent evidence is consistently pointing out that the subjective experience of ART-related toxicities might be as an important parameter to monitor in ART programs besides the ‘objective’ changes 
. Self-reported physical and psychological symptoms were identified as strong and independent risk factors for subsequent treatment failure in a recent study 
. From an operational perspective, the patient's perception is probably an integral part in quantifying the (‘subjective’) severity of toxicities. Integrating this in therapeutic decisions could contribute towards greater adherence to proposed interventions and towards improvement in the quality of life. This might be especially true for toxicities like body changes, where clear inter-individual differences in perception could exist. The negative impact of perception of body changes on quality of life and adherence has been reported in a number of studies 
A number of limitations have to be mentioned. This is a retrospective analysis, using data from a treatment program setting. Data on D4T dosing were not recorded in the database. However, given the overall low body weight in this population, few patients initiated high dose D4T (40 mg bid) before revision of the guidelines in 2006. D4T drug changes were driven by suspicion of D4T toxicity by the clinician, hence not necessarily with formal proof of D4T as the culprit drug. Reported data merely demonstrate associations, and not causation. Moreover, this program was probably better resourced with more intensive monitoring compared to other field settings, and this could possibly have resulted in over diagnosis of toxicities. With increasing experience of the program, diagnosis may have been influenced by negative perceptions about D4T by physicians and patients. However, patients were systematically evaluated at each clinical visit, with standard clinical assessment, patient management and reporting. We also note that data were collected prospectively using standardized data collection tools, with continuous monitoring of data quality and clinical management practices. Still, it remains that the unavailability of technical investigations to more rigorously diagnose the different toxicities could have led to misclassification. Moreover, at the start of the program, no lactic acid determination could be done. For lipoatrophy, diagnostic tools such as DEXA-scanning might have allowed to define the exact incidence of lipoatrophy over time. However, our main research question was not to define what happens ‘biologically’, but rather ‘operationally’ ie to what extent and for what reasons D4T is replaced over time within a program with close toxicity monitoring.
D4T-based treatment regimens in low-income countries are associated with significant long-term toxicities. With seven out of ten patients developing lipoatrophy by six years of treatment, lipoatrophy represents the major long-term side-effect. Until D4T has been phased-out completely, close monitoring for toxicity combined with the integration of the patient's perspective and a low threshold for D4T-replacement is recommended.