| Indicator | Data source(s) | Notes and assumptions |
|---|---|---|
| Mean time from diagnosis to viral suppression | 2013-2017 WA surveillance data | Viral suppression is defined as VL <200. This indicator is defined using data on MSM (MSM + MSM-IDU cases) diagnosed with HIV between 2013 and 2017 |
| Proportion of those diagnosed who are engaged in care | 2017 WA surveillance data | Engaged in care is defined as evidence of one or more HIV labs (CD4 or VL) in the past year, consistent with how the DOH defines engaged in care for their HIV care cascade. This indicator is defined using data from all prevalent MSM/MSM-IDU cases in 2017 |
| Proportion of those diagnosed who are virally suppressed | 2017 WA surveillance data | Viral suppression is defined as having a viral load result in that year, with the most recent VL <200. This indicator is defined using data from all prevalent MSM/MSM-IDU cases in 2017 |
| Proportion of those diagnosed who are durably suppressed | 2017 WA surveillance data | Durable suppression is defined as having a suppressed viral load (<200 copies/ml) over two consecutive labs X-months apart. We define this using intervals of both 6- and 12-months and take the weighted average. We also account for the proportion of men who are out of care. This indicator is defined using data from all prevalent MSM/MSM-IDU cases who had a first suppressed viral load measure in 2016. |
| Proportion diagnosed with late-stage HIV | 2013-2017 WA surveillance data | This indicator is the proportion of MSM with an HIV diagnosis in 2013-2017 who were diagnosed with AIDS within 12 months of their HIV diagnosis |
| Mean time from infection to diagnosis | 2017 WHAMP survey | On average, infection is assumed to occur midway between testing events (aka mean time from infection to diagnosis is half the intertest interval estimated from WHPP. Here we take the average ITI across all ages) |
| Proportion of men who progress to AIDS each year of infection | Lodi et al. 2011, Clin Infect Dis | Used to calculate the proportion of late diagnoses that do not test regularly |
| Duration of untreated asymptomatic infection | Lodi et al. 2011, Clin Infect Dis; Lee et al. 1991, BMJ | This is an estimate of how long someone would live before developing AIDS-associated sypmtoms in the absence of ART |
| Average time to viral suppression | Walmsley et al. 2013, NEJM | The average time from treatment initiation to viral suppression |
The model inputs that we calculate from these data include:
To represent heterogeneity by race/ethnicity and region in the care cascade, we estimate each of these parameters stratified by race/ethnicity and by region (with the exception of rates of treatment cessation). We then calculate the values for each racial/ethnic by region combination assuming independence.
To estimate the proportion of men who do not test for HIV, we use WADOH surveillance data on the proportion of MSM/MSM-IDU cases that are diagnosed as late-stage HIV, meaning they are diagnosed with AIDS within 1 year of their HIV diagnosis. However this does not correspond directly to the size of the non-regular tester group since some people’s CD4 count drops below 200 cells/mm^3 early in the course of infection (~8.8% in the first year of infection and 12.2% by 2 years4). As such, some of these late diagnoses might be regular testers - there is a lower bound for late diagnosis that we can never drop below even if everyone tests regularly.
Since the average intertest interval for regular testers is 436 days5, we will assume that regular testers who acquire HIV are diagnosed on average 218 days after seroconversion. Late diagnosis is defined as diagnosis with AIDS within 1 year of diagnosis, so regular testers would be considered late diagnoses if they progressed to AIDS within 583.25 days (~1.6 years) of seroconversion. If we assume that 8.8% progress to AIDS within 1 year and 3.4% progress in the second year (for a cumulative 12.2% by year 2)6, 0.108293 of regular testers will be late diagnoses.
The observed proportion of diagnoses that are defined as late is a combination of men who would test before sypmtom onset (regular testers) and men who test only with symptom onset (never testers): obs.late = nev.test + (1-nev.test) x (0.108293). Solving for never testers, we get: nev.test = (obs.late - 0.108293)/(1 - 0.108293). Below, we solve this equation using the estimates of the proportion of diagnoses that are late for each racial/ethnic group and region. Because we do not represent heterogeneity in intertest interval by race/ethnicity or region, we assume the proportion of regular testers who are diagnosed late to be the same across these groups.
| Region | Race/ethnicity | Estimate |
|---|---|---|
| King County | Black | 0.0803 |
| King County | Hispanic | 0.0638 |
| King County | Other | 0.1042 |
| Western WA | Black | 0.1204 |
| Western WA | Hispanic | 0.1015 |
| Western WA | Other | 0.1477 |
| Eastern WA | Black | 0.1614 |
| Eastern WA | Hispanic | 0.1402 |
| Eastern WA | Other | 0.1916 |
Because the date of ART initiation is not recorded in surveillance data7, we use data on median time from diagnosis to viral suppression to estimate time from diagnosis to ART initiation. A limitation of this approach is that the recorded date of viral suppression does not necessarily correspond to when viral suppression was first achieved; it depends on the timing of labs. As such, based on expert consultation with Drs. Matt Golden and Julie Dombrowski, we decided to assume that men initiate ART an average of 8 weeks before the recorded date of viral suppression. For the overall population, this implies a median time from diagnosis to ART initiation of 7.14 weeks, which aligns with Matt Golden’s intiution for the average time to initiation8.
While using the median to calculate a fixed time to treatment initiation does not reflect the heterogeneity in time to treatment initiation, we lack sufficient information to define a probability distribution. One option we considered is to use data on both the mean and median times to viral suppression, subtract 8 weeeks from each, and sample from a lognormal distribution with parameters \(\mu\), which can be expressed as \(ln(median)\), and \(\sigma^2\), which can be expressed as \(2*ln(mean/median)\) (see issue #93). However, the mean observed time to viral suppression is likely influenced by people who take much longer than average to acheive suppression and likely started treatment more than 8 weeks preceding their first suppressed viral load measure. For alternative ideas and notes, see file “/homes/dpwhite/R/GitHub_Repos/WHAMP/DWR_dissertation_Parameter estimation/Notes and draft files/test_method_for_sampling_time_to_ART_initiation.R” and issue #97.
Because this approach only uses data from men who achieve viral suppression (~4% do not have a recorded time to viral suppression), we assume that time from diagnosis to treatment initiation is the same for men who achieve suppression as for those who don’t. In reality, men who never achieve suppression likely take longer to initiate treatment, but we don’t have data to inform this. More research to determine the implications of this parameter and identify better data to inform it are needed.
| Region | Race/ethnicity | Median |
|---|---|---|
| King County | Black | 46.4 |
| King County | Hispanic | 43.3 |
| King County | Other | 46.7 |
| Western WA | Black | 56.3 |
| Western WA | Hispanic | 53.3 |
| Western WA | Other | 56.6 |
| Eastern WA | Black | 53.0 |
| Eastern WA | Hispanic | 50.0 |
| Eastern WA | Other | 53.4 |
We use data on the proportion of diagnosed men who are durably suppressed and the proportion out of care to estimate the probability of treatment cessation per weekly time step among full suppressors. This calculation assumes that the process of treatment cessation follows a geometric distribution, and requires that there was a set interval (X) over which durable suppression is measured. With these conditions met, we calculate the geometric distribution parameter p that has a CDF of Y at X, where \(Y = (1-\mathrm{P}(ds))\), and X = the months between the initial suppressed and follow-up lab visits. The CDF of the geometric distribution is \(1 - (1-p)^X\). Solving for p, we get \(1-(1-Y)^{1/X}\).
A challenge with surveillance data on viral suppression is that there are a wide range of intervals between lab records. To address this, we define a sample that includes only men who were initially suppressed (VL <200 copies/ml) at their first visit in 2016 and whose next visit was within X +/- 1 months. Standard intervals for labs are 6 and 12 months: men who have been on treatment for a while and are managing their treatment well might be told to come back at 12-month intervals, whereas men newly on ART or who are having more difficulty with adherence will come in biannually. As such, using either the 6- or the 12-month intervals might bias our estimates. We therefore use both intervals and take the weighted average of the calculated rates. A second issue is that, in defining the sample for this analysis to include only men who had a second viral load measure in the defined intervals, we will over-estimate durable suppresion because men who are out of care are not included in the denominator. So we define the probability of remaining durably suppressed through X months as the probability of being in care times the probability of remaining suppressed given that you’re in care: \(\mathrm{P}(ds) = \mathrm{P}(IC) \cdot \mathrm{P}(ds \mid IC)\). We define the probability of being in care as 1 minus the probability of being out of care, which we measure using data on the proportion of men who were suppressed at their first visit in 2016 who had no labs for the subsequent 18 months and who are not on record as having died or moved to another jurisdiction. For more on the rationale for this method, see GitHub issue #81.
Due to small numbers in some racial/ethnic and regional groups that limited our ability to obtain stratified estimates for these inputs, we assume the rate of treatment cessation is homogenous for all men in Washington. We set the probability of treatment cessation for partial suppressors to be twice that of full suppressors (see section @ref(decisions)). As described below, we calculate differential rates of treatment reinitiation by racial/ethnic and regional groups such that the proportion of men on treatment in the cross-section matches observed estimates.
| Probability | |
|---|---|
| Full suppressors | 0.002 |
| Partial suppressors | 0.004 |
We use the calculated rate of treatment cessation for full suppressors and observed proportions of diagnosed men on treatment and treated men with viral suppression to solve for a) the rates of treatment reinitiation for full and partial suppressors that are consistent with the observed proportions on treatment, and b) the proportions of men who treat with full suppression. This latter estimate will not be the same as the observed proportion with viral suppression because it accounts for the fact that, in the cross-section, some full suppressors and some partial suppressors are off treatment. We assume that survival is equivalent for full and partial suppressors (see section @ref(decisions)), all diagnosed men initiate treatment, and the proportion of men who treat with full suppression is the same for regular testers and for men who are diagnosed only upon progression to symptomatic HIV.
| Region | Race/ethnicity | Full suppressors | Partial suppressors |
|---|---|---|---|
| King County | Black | 0.0244 | 0.0122 |
| King County | Hispanic | 0.0239 | 0.0120 |
| King County | Other | 0.0275 | 0.0138 |
| Western WA | Black | 0.0200 | 0.0100 |
| Western WA | Hispanic | 0.0195 | 0.0098 |
| Western WA | Other | 0.0223 | 0.0112 |
| Eastern WA | Black | 0.0219 | 0.0110 |
| Eastern WA | Hispanic | 0.0209 | 0.0105 |
| Eastern WA | Other | 0.0237 | 0.0118 |
| Region | Race/ethnicity | Proportion |
|---|---|---|
| King County | Black | 0.839 |
| King County | Hispanic | 0.902 |
| King County | Other | 0.920 |
| Western WA | Black | 0.811 |
| Western WA | Hispanic | 0.885 |
| Western WA | Other | 0.905 |
| Eastern WA | Black | 0.759 |
| Eastern WA | Hispanic | 0.847 |
| Eastern WA | Other | 0.874 |
shared via email by Steven Erly on 12/6/18↩
Lodi et al. 2011, Clin Infect Dis; Lee et al. 1991, BMJ↩
Walmsley et al. 2013, NEJM↩
Lodi et al. 2011, Clin Infect Dis↩
Note we exclude PrEP users because PrEP users have different testing patterns and are unlikely to acquire HIV↩
Lodi et al. 2011, Clin Infect Dis↩
A commonly reported indicator in surveillance data is the time from diagnosis to “linkage to care,” with a case considered “linked” at the time of first CD4 or viral load lab records. In the 2018 Washington State and King County HIV/AIDS Epidemiology Report, 91% of MSM living with diagnosed HIV in King County were linked to care within 1 month of diagnosis and 97% were linked within 3 months. Linkage to care does not indicate ART initiation, as some people require additional visits before initiating, but it provides information to gauge how realistic our calculations of time to ART initiation are: they should not be wildly off from this.↩
This intuition is based on observations in King County, but since the median time to viral suppression is fairly similar across the state, we assume that this method results in an accurate approximation for the entire state↩