It appears that the truncate.plist control setting, which has a default of TRUE, has not been modified (set to FALSE) during the testing and validation so far. Is that correct?

In the current workflow, it is possible that an identified partner could be identified in more than one partnership:

p1 p2 ptype start stop
1   5    2        10    NA
72 5    2        20    50

In this scenario 1 and 75 have tested positive and correctly recall individual 5. Should this be treated (as it is now) as one testing event, with two chances for identification or should everything be done at the partnership level?

@smjenness

In hivtx_msm we wanted to allow individuals who have been identified through the elicitation process to possibly reinitialize ART services at a higher rate. Is this meant to only occur at the time they are identified, or, like halting of ART services, meant to be ongoing and based only on the part.art parameter (identifies if someone has been on ART's and identified through the identification process).

  ## Restarting (partner services)
  part.tx.reinit.elig <- which(tx.status == 0 & part.art == 1 & part.ident == at)

Currently, partner identification and screening for HIV occurs within hivtest_msm. Instead, we would like to pull out the identification process into it's own separate module (part.ident) and update hivtest_msm to account for individuals who are testing during a general testing process and those who are testing due to partner notification.

It is not a logical, as it is used here, it is a time step

https://github.com/EpiModel/CombPrevNet/blob/master/test/03-validation_PartnerPrEP.R#L23

Currently in EpiModel, the plist is updated at every step. In particular, the time at which a relationship ends will be recorded based on results of a call to tergmLite::simulate_network. Unless specified otherwise, code comes from here:
https://github.com/EpiModel/EpiModelHIV-p/blob/CombPrev/R/mod.simnet.R

dat$el[[1]] <- tergmLite::simulate_network(p = dat$p[[1]],
                                             el = dat$el[[1]],
                                             coef.form = nwparam.m$coef.form,
                                             coef.diss = nwparam.m$coef.diss$coef.adj,
                                             save.changes = TRUE)

plist1 <- update_plist(dat, at, ptype = 1)

Thus, the entries in plist that are assigned are based on the updated edgelist (specifically the "changes" attribute of the dat$el[[ptype]] object):

update_plist <- function(dat, at, ptype) {
  # pull existing partner type specific list
  plist1 <- dat$temp$plist[dat$temp$plist[, "ptype"] == ptype, ]

  # look up dissolutions, update stop time
  uid <- dat$attr$uid
  news <- attr(dat$el[[ptype]], "changes")
  news_uid <- cbind(matrix(uid[news[, 1:2]], ncol = 2), news[, 3])
  news_uid_stop <- news_uid[news_uid[, 3] == 0, , drop = FALSE]
  pid_plist1 <- plist1[, 1]*1e7 + plist1[, 2]
  pid_stop <- news_uid_stop[, 1]*1e7 + news_uid_stop[, 2]
  matches_stop <- match(pid_stop, pid_plist1)
  plist1[matches_stop, "stop"] <- at

  # look up new formations, row bind them
  news_uid_start <- news_uid[news_uid[, 3] == 1, , drop = FALSE]
  plist1 <- rbind(plist1, cbind(news_uid_start[, 1:2, drop = FALSE], ptype, at, NA))

  return(plist1)
}

Then, the plist is updated latter on by dropping those who's relationships have stopped:

  if (dat$control$truncate.plist == TRUE) {
    to.keep <- which(is.na(dat$temp$plist[, "stop"]))
    dat$temp$plist <- dat$temp$plist[to.keep, ]
  }

However, this method for dropping individuals from the plist fails to drop relationships that are ended by an ego exiting the population (I think). Code in the mod.departure script drops edges:

  if (length(idsDepAll) > 0) {
    dat$attr$active[idsDepAll] <- 0
    for (i in 1:3) {
      dat$el[[i]] <- tergmLite::delete_vertices(dat$el[[i]], idsDepAll)
    }
    dat$attr <- deleteAttr(dat$attr, idsDepAll)
    if (unique(sapply(dat$attr, length)) != attributes(dat$el[[1]])$n) {
      stop("mismatch between el and attr length in departures mod")
    }
  }

However, I don't think these changes are stored in the dat$el[[ptype]] changes attribute as a result, these relationships stored in the plist will never be assigned a date for the relationship ending, and thus the will never be dropped from the plist.

I am not sure how important this is to anyone else, or how clear this is. If anyone has questions, just let me know. Thanks!

Translate word document into wiki page outlining schematic.

Add a parameter to allow for random screening of partners identified by recently diagnosed HIV positive individuals.

With the introduction of a new partner identification module, #23, wiki pages will need to be updated accounting for this change.

parameter name: part.index.degree
default: 1
function: this would limit index patients in the cumulative partner list to those with a degree of part.index.degree or greater. For example, part.index.degree = 3, would remove from the list any index patients who show up in the list only 1 or 2 times.

Identification of partners for testing should not be fixed; instead introduce identification probability parameter based on partnership type; e.g. part.ident = c(0.75, 0.5, 0.1).

Along with checks on signal output, there should be a check on internal consistency of simulation results. As an example part.identified, the attribute tracking total number of partners identified within some timing window, should be equal to or less than part.screened, the attribute tracking whether an identified partner screened or not.

All partners who screen negative during partner based screening - HIV never tested, screen negative at time at and previously screened HIV negative individuals - are then started on PreP. This will be an update of the module for starting of prep: prep.net.

For example: lines 112 and 148

Update R to 4.x
Update all packages to current.
Use current CRAN version of statnet packages
Use current CRAN version of tergmLite
Use Github master of EpiModel

Update library, then run renv::snapshot(), then commit the renv.lock file

This is unclear:

#' @param recent.diag The number of time steps back that a HIV+ MSM may be considered
#'        to be incident for purposes of partner elicitation.

Partner lookback, the time interval used to identify partners of newly HIV diagnosed individuals, should be flexible and based initially on partnership type.

Currently stratified by full, partial, durable

With new HIV+ identified through partner process (either through screening, or previous HIV diagnosis) we want to add the option that potentially individuals are engaged in supportive HIV care services, e.g. a reduction in dropout.

This should be done whenever new functionality is developed for the project:

• internal test to see the model runs without error
• run code in debug model to catch any "silent errors"
• run simulations across a range of parameter values, including the extreme parameter values (e.g., probability parameters of 0 and 1)
• demonstrate parameter signal by showing relationship between inputs and outputs

Following updates to ergm and tergm, errors now introduced during network resimulation related to age.group missing data.

Addition of the attribute time.ident to partners identified for testing at time step at. At the end of testing, update attribute to denote when partner is identified, to make sure we are not duplicating testing of individuals within the "lookback period" for partner identification. Once at - time.ident is greater than partner lookback period, reset this to NA to allow for reentry into partner eligible testing.

Incorporate a parameter that allows for partner identification to be based on partnership type: e.g. ptype.lookback = c(1, 2, 3) or any combination of the three partnership types for partner notification.

One-off partnerships that occur at timestep at = t, need to have both their start and stop times set to at.

In EpiModelHIV, initiation of PreP services is done when the parameter prep.start has passed a predefined step. For CombPrevNet, we want to include a separate parameter that handles initiation of PreP services for partners identified: prep.start.part.

As detailed here, under Discussion, if the partner lookback period is too high, MSM identified through partner elicitation are more likely to have had a recent test than if the lookback period was shorter. This is because eligibility for testing is tied up in partner identification, when - logically - the two should be separate: individuals are identified regardless of testing status of the identified partner; it is only in hivtest_msm that an individuals testing elig. should be evaluated. However, because this eligibility depends on partnership type, this is calculated at the same time when calculations are done using the temporary parntership list.

This has the effect of dampening total partners identified and partners tested, when it should in fact dampen only partners tested.

https://github.com/EpiModel/CombPrevNet/blob/master/R/00-fx.R#L40-L52

It seems like ptype.lookup restriction is unnecessary if the partner.ident.XXX are 0. Clarify purpose here.

Update all partner based parameters to start with part.

Time since last test, tsincelntst, is only pulled at beginning of hivtest_msm; it needs to be updated at each stage of HIV testing prior to partner testing.

Current version just subsets index patients who test positive on current time step

Update documentation for EpiModelHIV@CombPrevNet.

https://github.com/EpiModel/EpiModelHIV-p/blob/CombPrevNet/R/mod.partident.R

Addition of a Wikipedia page detailing steps for initiation of prep for HIV negative screened partners.

In code below, hiv.scrn.rate is nowhere in EpiModelHIV

param <- param_msm(epistats = epistats,
                   netstats = netstats,
                   hiv.scrn.rate = c(1, 1, 1),
                   part.identification = 1,
                   part.lookback.main = 52,
                   part.lookback.casl = 52,
                   part.lookback.ooff = 52,
                   part.ident.main = 1,
                   part.ident.casl = 1,
                   part.ident.ooff = 1,
                   ptype.lookup = c(1, 2, 3)
)

In my two recent commits on EpiModelHIV:

1. ed9af66061d9cfe54867efb454e48a7b80ea5bd2
2. 1c72dc069fdc60c5b85b8d6cef738662b47261e8

I fixed when diag.time was greater than present in the init status fx, and then the logic for recent positive diagnosis selection in the partner ident fx.

Please double-check this makes sense.

Instead of TRUE/FALSE, it would retain all partnerships that have stopped within X number of time steps. For example, if X was 12, then we would keep all active partnerships and those that have ended within the past 12 time steps. This will improve model efficiency.

Leave open until project complete.

Introduce a new attribute so that we are keeping track of people who are eligible for ART services through partner identification: we want to make sure that if someone screens negative through partner identification, and then subsequently screens positive during general testing - that they are not onboarding to ART (or PreP is general screening negative) at an elevated rate.

As discussed on Slack

      dat$attr$part.ident <- ifelse(tsincelstident >= part.ident.window[race], NA,
                                    dat$attr$part.ident)

Create wikipedia page for schematic breakdown of linking HIV+ partners to care through hivtx_msm. Note: use work on Model Schematic 2 as a guide for this process.

The below will run through testing of each stage of the CombPrevNet workflow in EpiModelHIV.

For halting, the process is split into PN and individual-level processes:

  ## Halting: Identified through partner elicitation
  tx.halt.part.elig <- which(tx.status == 1 & tt.traj == 1 & part.art == 1)
  rates.part <- part.tx.halt.part.prob[race[tx.halt.part.elig]]
  tx.halt.part <- tx.halt.part.elig[rbinom(length(tx.halt.part.elig), 1, rates.part) == 1]

  tx.halt.full.elig <- which(tx.status == 1 & tt.traj == 2 & part.art == 1)
  rates.full <- part.tx.halt.part.prob[race[tx.halt.full.elig]] * tx.halt.full.rr[race[tx.halt.full.elig]]
  tx.halt.full <- tx.halt.full.elig[rbinom(length(tx.halt.full.elig), 1, rates.full) == 1]

  tx.halt.dur.elig <- which(tx.status == 1 & tt.traj == 3 & part.art == 1)
  rates.dur <- part.tx.halt.part.prob[race[tx.halt.dur.elig]] * tx.halt.dur.rr[race[tx.halt.dur.elig]]
  tx.halt.dur <- tx.halt.dur.elig[rbinom(length(tx.halt.dur.elig), 1, rates.dur) == 1]

  tx.halt <- c(tx.halt.part, tx.halt.full, tx.halt.dur)

  ## Halting
  tx.halt.part.elig <- which(tx.status == 1 & tt.traj == 1)
  rates.part <- tx.halt.part.prob[race[tx.halt.part.elig]]
  tx.halt.part <- tx.halt.part.elig[rbinom(length(tx.halt.part.elig), 1, rates.part) == 1]

  tx.halt.full.elig <- which(tx.status == 1 & tt.traj == 2)
  rates.full <- tx.halt.part.prob[race[tx.halt.full.elig]] * tx.halt.full.rr[race[tx.halt.full.elig]]
  tx.halt.full <- tx.halt.full.elig[rbinom(length(tx.halt.full.elig), 1, rates.full) == 1]

  tx.halt.dur.elig <- which(tx.status == 1 & tt.traj == 3)
  rates.dur <- tx.halt.part.prob[race[tx.halt.dur.elig]] * tx.halt.dur.rr[race[tx.halt.dur.elig]]
  tx.halt.dur <- tx.halt.dur.elig[rbinom(length(tx.halt.dur.elig), 1, rates.dur) == 1]

  tx.halt <- c(tx.halt.part, tx.halt.full, tx.halt.dur)

But the IDs that halt from the PN process would be overwritten by the individual-level process because you are using tx.halt to track both. Can you explain what is going on here?

Can ignore:

• snake_case variable names

With the addition of partnership level testing, current epitrackers need to be updated and new epitrackers introduced. Some examples: total testing (updated), newly diagnosed (updated), last negative test (updated), total partners identified (new), total partners screened for testing (new), etc.

This would wrap around this code:

https://github.com/EpiModel/EpiModelHIV-p/blob/5616ea8efe6760825720357a0cffab685a731e64/R/mod.initialize.R#L93-L100

And do a for loop for the number of networks: for (i in 1:length(dat$el))...

Put this and the update_plist functions together in a separate R file in R/.

It looks like tergmLite is not being used in model calibration:

control <- control_msm(simno = fsimno,
                       nsteps = 52*60,
                       nsims = ncores,
                       ncores = ncores,
                       save.nwstats = FALSE,
                       save.clin.hist = FALSE,
                       tergmLite = FALSE)