ART discontinuation

[darcyrao]

• Ruanne says people should move back to the CD4 and viral load state they were in at the time they started ART. Since it is not an individual model, we can't do this exactly, but could allocate people according to the distribution among people who started on ART over some number of time steps.

[darcyrao]

Rate of discontinuation:

• A study by Moosa et al. 2019 in Durban, South Africa with 270 patients recruited 2005-2008 reported a median time to ART discontinuation of 70 months (IQR: 64-78). Mean adherence was >=95%. At the end of the study, 94% were virally suppressed, and females were more likely to have high adherence. Adherence increased over time from ART initiation. Cumulative 10.7% probability of treatment failure at 5 years from initiation, mostly in the first two years.

  • Using the exponential distribution formula CDF = 1-e^(-rx) to translate estimates of the proportion of women who discontinued by x months to a rate, the rate = -ln(1-CDF)/x. For a median discontinuation of 70 months, this translates to an annual rate of 11.88%.
  • Although it was an RCT that included adherence support and follow-up with no-shows such that patients may not be generalizable, the data support assuming that persons on ART have viral suppression.

• A 2011 study of women initiating ART in Johannesburg, SA (El-Khatib et al. 2011) reported that 37% of women had incomplete adherence (>5% of pills returned at any visit) in the first 24 weeks from initiating ART and 7% had VL >400/ml at 24 weeks. But these women all had fairly advanced disease at enrollment, many were single mothers. did receive adherence support as part of the trial though.

• A 2010 study in Khayelitsha township in South Africa (Ford et al. 2010 reported that 87% of patients were highly adherence (>95%) in the first months after ART initiation. Among patients with high adherence, ~10% had virological failure at 2 years and ~17% at 5 years. Among those with adherence <95%, ~32% had virological failure at 2 years and ~38% at 5 years. Virological failure (two consecutive viral loads >5000 copies/ml) can be due to suboptimal adherence, poor drug potency, and/or drug resistance.

• Orrell et al. 2003 conducted a cohort study in Cape Town and found that adherence to ART is high. 16.2% of patients discontinued ART within 48 weeks. Median adherence up to 48 weeks was 93.5% , including 8 patients who had 0% adherence). 66.1% of those still on ART at 48 weeks had viral load <400 copies/ml (73% among those with high adherence)

  • This indicates that only 2/3 of those on ART have viral suppression, although this might change with more effective regimens and earlier start. Patients were recruited between 1996 and 2001.
  • Annual rate of discontinuation = 19.2% using the above exponential distribution formula

• A meta-analysis in 2006 (Mills et al. 2006) reported higher combined adherence in South Africa than North America -- pooled estimates 77% vs. 55%).

• A prospective cohort study in rural KZN South Africa from 2012-2016 (Iwuji et al. 2018 found 73.5% of visits had adherence >=95% measured on both VAS and with pill counts, and higher if measured on either. Starting ART regardless of CD4 count (vs. starting with CD4 <350) did not impact adherence. 97% of participatns with viral load data at 12 months were virally suppressed. Viral suppression was high even in those with <95% adherence at >90%.

  • Adherence ~75%. Higher in women.
  • But very high viral suppression -- >90% even in those with suboptimal adherence.

• Orrell et al. 2017 compared different measures of adherence over 48 weeks in a cohort of PLWHIV in Cape Town. They were all starting first-line ART. Part of an RCT assessing the impact of phone message reminders on adherence, so results may not be generalizable. At 48 weeks, 8.3% were lost to follow-up. By week 16, 91% had viral suppression (of those with viral load measured) and at week 48 85% had viral suppression. By wisepill, median adherence was 93% and 86% at weeks 16 and 48. 92% had therapeutic drug levels at 48 weeks.

  • Data suggest a discontinuation rate of 9.4% per year
  • Adherence of ~90% at 48 weeks and 85% viral suppression, although it is part of an RCT so may not be generalizable.

• Mills et al. 2011 report 6.4% loss to follow-up over a median of 31 months of follow-up. But they used mobile teams to track patients to increase retention, so results may not be generalizable.

• A systematic review of ART retention in LMICs from 2008-2013 (Fox and Rosen 2015) reported pooled estimates of retention at 83%, 74%, 68%, 64%, and 60% for 12, 24, 36, 48, and 60 months. Retention was lower in Africa at 65% at 36 months. These estimates don't account for transfers to other clinics, so the percent attrition reflects deaths (~43% of LTFU) as well as discontinuation and transfers. Their previous reviews for sub-Saharan Africa estimated 62% retention at 24 months prior to 2007 and 76% between 2007-2010 (73% at 36 months) -- their analysis of temporal trends in this updated review suggests that retention was slightly higher in studies that started enrollment before vs. after 2008.

• Wilkenson et al. 2015 conducted a systematic review and meta-analysis of data on patients LTFU in LMICS (all studies were in sub-Saharan Africa). They estimate that 18.6% of patients LTFU had transferred care to a different facility.

• Sikazwe et al. 2019 traced outcomes for patients lost to follow-up to assess the impact on estimates of retention in care in Zambia. Over 24 months, 17% of patients in 64 facilities in Zambia were lost to follow-up. They randomly selected 16% of cases lost to follow-up to trace and found that 14% had died. Of those alive, 74% were still on treatment (including 48% still in care at original facility). Crude estimate of retention using EMR data was 72.3% at 2 years, but after tracing increased to 91.2%. But viremia was significantly higher in patients lost to follow-up from one facility to received care in a new facility than those who remained in the original facility.

• Chammartin et al. 2018 conducted a meta-analysis of outcomes among patients lost to follow-up for HIV care in sub-Saharan Africa. Include data from 9 studies with 7377 patients lost to follow-up in SSA. Most patients were lost in the first 6 months from initiation of ART, with a median time before LTFU of 162 days. Tracing indicated that 21.8% of patients LTFU had died, 22.6% had stopped ART but were alive, 14.8% had transferred, 9.2% remained on ART at their original clinic, and 31.6% could not be found.

• Geng et al. 2011 measured retention and traced cases LTFU in a clinic in rural Uganda from 2004-2007. Retention estimates at 1, 2, and 3 years with and without tracing were 82.3%, 68.9%, 60.1%. With data on LTFU cases, estimates ranged from 85.8%-90.0%, 78.9%-86.2%, and 75.8-84.7% at 1, 2, and 3 years.

[darcyrao]

Update from meeting with Ruanne:

• Don't model discontinuation, at least in the near term. Having people churn on and off ART is problematic because it distributes the benefits of ART to basically all women, such that the coverage level in the scenarios is somewhat irrelevant. As Cara's diagnostic plots show, this pretty much eliminates the HIV epidemic by 2120 because the population-level CD4 increases and VL decreases. An alternative for future work would be to define a separate compartment for people who will go on ART and those who will not, and only cycle people on and off ART from the "will go on ART" compartment. For now, instead just adjust coverage to account for ~5% per year dropping off. Keep the same assumptions for viral suppression and CC multipliers for women on ART -- while in reality their CD4 might be higher and VL lower due to cycling on and off, we don't have data to inform that and it can be quite complicated.
  • Sources for the 5% per year:
  • [x] Look at Roger Ying's paper for references
  • [x] Read papers on loss to follow-up (added to above review)

[carajbro]

Ying. Lancet HIV. 2016. Home HIV testing and counseling for reducing HIV incidence in a generalized epidemic setting: a mathematical modeling analysis "ART treatment is assumed to reduce the likelihood of HIV transmission by 96% as suggested by recent studies, and persons on ART are expected to have the same life expectancy as HIV-negative persons of similar age and gender, and thus, are assumed not to be subject to HIV-associated mortality. The annual drop-out rate is 6%, which is equally likely for all individuals regardless of their HIV state prior to treatment. Individuals who drop out of ART return to the infected stages at the same proportion with which they enrolled."

SOURCES REFERENCED Mills. AIDS. 2011. Mortality by baseline CD4 cell count among HIV patients initiating ART: evidence from a large cohort in Uganda -6.4% patients lost to follow-up. Cohen. N Engl J Med. 2011. Prevention of HIV-1 Infection with Early ART -I believe this is where the 0.04 reduction multiplier for HIV transmission comes from. Attia. AIDS. 2009. Sexual transmission of HIV according to VL and ART: systematic review and meta=analysis Donnell. Lancet. 2010. Heterosexual HIV-1 transmission after initiation of ART: a prospective cohort analysis Jahn. Lancet. 2008. Population-level effect of HIV on adult mortality and early evidence of reversal after introduction of ART in Malawi

[darcyrao]

Added Mills study data to above review, but in the discussion they comment that their loss to follow-up is lower than many other sub-Saharan African sites because they use a "mobile team on motorcycles that consistently track patients"!

[darcyrao]

SUMMARY:

• Studies report 10-20% attrition over 1 year, and systematic reviews suggest ~17% by one year. But this includes patients who died and those who transferred to other facilities. Fox and Rosen found that, in studies that disaggregated deaths from LTFU, ~43% of cases lost to follow-up died. In Zambia, Sikazwe et al. (2019) found that, after tracing outcomes for patients LTFU, retention at 2 years was ~91.2% and only 4% were out of care and alive. A meta-analysis by Chammartin et al. (2018) concluded that 23% of patients LTFU were alive and out of care, and 32% could not be found. Applying this to the 10-20% LTFU over one year suggests 2-11% of patients discontinue ART (but remain alive) per year (lower estimate assumes 10% LFTU and 23% out of care, upper estimate assumes 20% LTFU and that all those who could not be found are out of care).

So it seems reasonable to assume that 5% per year discontinue and remain alive.

[darcyrao]

Note: A DCM parameterized to South Africa by Eaton et al. incorporates discontinuation, with differential rates depending on time on ART and CD4 count at initiation. But to do this, they had to define many different ART compartments, which would not be very feasible for our model. The appendix with their detailed assumptions is here: file:///C:/Users/dpwhite/Downloads/pnas.201323007SI.pdf. See section1.5 and table S9.