Authors’ Response

Sir,

We thank the authors for their correspondence¹. We appreciate this opportunity not only to reinforce for the journal audience how our findings were interpreted in full consideration of the inherent limitations of cross-sectional survey methodology, which we explicitly outlined in our paper², but also to expand upon the pertinent issues raised.

As detailed in our article (p. 254) and Supplementary Material 2, we had already discussed the relevance and limitations of using data from 2014/15, including the potential impact of changes in exposures and outcomes over time. However, available evidence suggests that the structural and policy landscape for human immunodeficiency virus (HIV) prevention among people who inject drugs in India has remained essentially unchanged. National surveillance data indicate that HIV prevalence among this population has stayed high and relatively stable (9.9% in IBBS 2014/15² vs. 9% in HSS-Plus 2021³), and opioid substitution therapy coverage remains limited (14.9% in HSS-Plus 2021³). Structural barriers to accessing healthcare services due to stigma, including fear of violence, are still reported by people who inject drugs at higher levels (24.4% in 2021³). Needle and syringe programmes under NACO’s targeted HIV interventions may have contributed to the national decline in needle/syringe sharing among people who inject drugs (national average: 17.8% in IBBS 2014/15² vs. 4.2% in HSS-Plus 2021³); however, sharing remains substantial in certain regions, for example, 36.8 per cent in Arunachal Pradesh in 2021³. Importantly, needle/syringe programmes are still not available in prisons, the incarceration rate of drug users (who used drugs for personal consumption) is reported to have increased from 2017 to 2021⁴, and opioid substitution therapy is not universally accessible in these settings, except certain prisons like Tihar⁵. Thus, while some behavioural indicators have improved, key structural, legal and policy contexts have not changed significantly since the IBBS 2014/15 survey. Accordingly, the main findings and implications of our study remain relevant to current programmatic and policy efforts. Nonetheless, we agree that periodic, comprehensive studies are needed to monitor trends and inform future interventions.

We agree with Dr. Nagar that the inability of our analysis to shed light on temporal relationships is a generic limitation of all cross-sectional studies, which we noted in our article (p. 254 and Supplementary Material 2). While the IBBS’s cross-sectional design limits definitive temporal ordering, the differing recall periods are relevant. Participants were asked about incarceration experience related to drug use ‘in the last 12 months’ (p. 250). In contrast, needle/syringe sharing was assessed with respect to the last time a study participant injected drugs. Given that injection drug use within the previous three months was one of the study inclusion criteria, the recall period for this variable would therefore typically be ‘within the previous three months’ (Supplementary Material 2). This difference in recall periods suggests that for a substantial proportion of participants reporting both exposures, the 12-month recall window for incarceration would likely encompass or precede the more recent timeframe for the reported needle/syringe sharing episode. Indeed, this very difference in measurement timeframes explains why we could not test a mutually causal model as stated in our article (Supplementary Material 2).

Recall bias and social desirability bias could yield underestimates of the prevalence of needle sharing and violence victimization, which we discussed in our article (Supplementary Material 2). We, however, disagree with Dr. Nagar that such biases would necessarily attenuate our estimates of the associations between these exposures and the outcome. Such biases could result in an upward bias of our estimates (i.e., away from the null). The direction of the bias is unpredictable.

As described in our article, the operationalisation of ‘severe violence victimization’ was a pragmatic choice based on the available response options to the question. Importantly, when we specified both violence victimisation and needle/syringe sharing as continuous variables as part of sensitivity analysis, our findings remained qualitatively similar (p. 253), suggesting that alternative operationalisations would not have altered our study conclusions.

We reported in our article that the study’s focus on men who inject drugs limits our ability to generalise the findings to women who inject drugs (p. 254), which we expanded in the Supplementary Material 2 also stated was ‘…another limitation is the exclusion of women who inject drugs from the original IBBS study. This exclusion limits our ability to examine gender differences in syndemic exposures and their impacts on HIV risk’, and we recommended that ‘future studies should prioritise including both men and women’.

As we noted in our article (Supplementary Material 2), we agree with Dr. Nagar that some subgroups of men who inject drugs might not frequent the venues used for time-location sampling in the IBBS, which limits our ability to generalise the findings of all subgroups of men who inject drugs in India.

The possibility for confounding by unmeasured variables is a generic limitation of all estimates derived from non-experimental data. We disagree with Dr Nagar that socioeconomic deprivation is likely to be an important confounder, given the well-known correlation between educational attainment and socioeconomic deprivation in the Indian context⁶, given that our regression models included adjustment for educational attainment, a proxy for socioeconomic deprivation. We agree that unmeasured psychiatric conditions could have potentially confounded our associational estimates.

We agree with Dr. Nagar that the inclusion of HIV programme exposure as a covariate does not assess its potential moderating effects on the identified syndemic relationships.

We concur with Dr. Nagar’s comment that readers unfamiliar with statistical methods may have difficulty interpreting the semi-elasticities. The footnote of the Table is likely to be helpful in this regard (p. 253): ‘The estimated semi-elasticities are interpreted as the percent relative change in the expected value of the outcome (HIV-positive status) that is associated with the interaction...’

While we agree with Dr. Nagar that small proportion (0.3%) of the sample experienced all three exposures (incarceration, violence victimisation, and needle/syringe sharing), which we explicitly stated in our article (p. 254), it is unclear to us why a statistically significant triple interaction would raise ‘concerns about statistical power and the stability of estimates.’

Ethical vigilance is paramount in research involving criminalised and vulnerable populations, particularly in light of the evolving legal and policy environments, as noted by Dr. Nagar. The IBBS study, conducted in 2014-15, adhered rigorously to the ethical standards of that time, with the study protocol receiving comprehensive ethical approval from the Ethics Committee and Technical Committee of the National AIDS Control Organization (NACO), as well as from the institutional ethics committees of all implementing research institutions across India, in line with the prevailing ICMR national ethical guidelines. It is unclear to us what specific ethical considerations Dr. Nagar believes must be revisited.

In conclusion, we are grateful for the discussion prompted by Dr. Nagar’s correspondence. We hope our responses reinforce the robustness of our conclusions, which were drawn with full awareness of the study’s limitations, while also illuminating the critical issues raised.