Read on Twitter
Our paper, RCT on PSH for the highest user of healthcare is out @HSR_HRET. PSH w/ intensive case management ends homelessness for 86% of the highest users of multiple services. A lot of subtlety here. A 
@mravenEM @ucsfcvp @ucsfbhhi @ucsf https://onlinelibrary.wiley.com/doi/full/10.1111/1475-6773.13553 1/x
@mravenEM @ucsfcvp @ucsfbhhi @ucsf https://onlinelibrary.wiley.com/doi/full/10.1111/1475-6773.13553 1/x
Multi-year RCT comparing PSH (subsidized housing, mostly scattered site; ICM 1:15 master’s trained led team by @abode_services) vs usual care for chronically homeless ppl who were highest users of acute healthcare/jail in @sccgov 423 participants (199 intervent/224 control). 2x
Inclusions: chronic homelessness; highest users of services. Ineligible: engagement other program, req nursing home, hospice eligible. We created risk score/triage tool to calculate highest users (mult criteria). Triage tool calculated continuously & flagged in records. 3/x
When highest users presented (flagged in system), staff alerted. County staff approached to explain study and determine eligibility. Approached 426; 2 refused to engage to determine eligibility & 1 eligible declined. Basically—everyone participated. 4/x
Ppts randomized immediately after consent: enrollment in Project Welcome Home or usual care. Those in usual care offered resource lists and remained eligible for other services. Why randomization? We had fixed # units in this program (Paid for in Pay for Success program). 5/x
We couldn’t distinguish bw group >2x size of available units. It was going to be random who got in anyway—knew doing RCT would let us answer some important ?s. Most studies of PSH (except chez soi, not yet published when started) do pre-post evals 6/x https://www.homelesshub.ca/solutions/housing-first/homechez-soi
Pre-post evaluations don’t account for regression to the mean. Also—may be selection bias in who gets into housing. We wanted to examine everyone randomized—whether or not they enrolled. Interested in reach & outcomes. 7/x
(Yes—RCTs typically don’t give info on reach—but all who qualified approached, all but 3 agreed to assessment/enrollment. So, we feel we can talk about reach). We followed outcomes by tracking data in systems (not in person interviews) What did we find? 8/x
We selected VERY complex ppl who were very high users: 2 yrs prior to enrollment, avg 5.1 inpatient admits, 19 ED visits, 3.7 jail stays, 36.7 shelter days. Per person. These individuals had high levels of behavioral health & medical complexity. And Chronic homelessness. 9/x
Btw July 2015-Sept 2019: 86% of those in intervention housed: 36% control group. (PSH 
during study period in @SCCgov d/t incredible work @DSTNhome @jenloving23). On avg, 74d from enrollment to housing (doesn’t happen on day 1!) 10/x
during study period in @SCCgov d/t incredible work @DSTNhome @jenloving23). On avg, 74d from enrollment to housing (doesn’t happen on day 1!) 10/x
On avg, those housed moved 2x during study (range 1-10). Those housed remained housed w/o gaps into homelessness for 93% study period (avg 29 months). Meaning: Not only were these folks housed, they stayed house, but needed flexibility when 1 place didn’t work. 11/x
Put another way, intvn group housed 84% of time vs 20% of control group—and most those, bc new, similar programs opened up. (Psst: PSH works—really well, we just need more). Shelter stays decreased (although most unsheltered & harder to measure). 12/x
Only stat diff in health care utilization: Intv outpt mental health (37 vs 20/year) and 
psych ED (1.3 vs 1.9/yr). Intention similar phys ED visits, inpt (med & psych) & jail. Why? Don’t many studies say PSH decreases utilization? 13/x
outpt mental health (37 vs 20/year) and psych ED (1.3 vs 1.9/yr). Intention similar phys ED visits, inpt (med & psych) & jail. Why? Don’t many studies say PSH decreases utilization? 13/x
Yes & no. Most studies not controlled, huge regression to the mean. (We saw in both groups). These ppl were v sick—hard to utilization at this stage—much may be appropriate. Once this sick, may be too late to utilization. Possible more with time. 14/x
utilization. Possible more with time. 14/x
B/c 36% of control group got PSH (yay), our ability to find subtle diff. Trend toward phys hospitalizations, maybe would have seen w more power, but large differences would have come through. 15/x
our ability to find subtle diff. Trend toward phys hospitalizations, maybe would have seen w more power, but large differences would have come through. 15/x
Is this disappointing? Not really—PSH ends homelessness. That is the goal, and it does it well. There are many reasons to do this, not just hope that it health care costs. Wrote this w @StefanKertesz @TPBaggett https://www.nejm.org/doi/10.1056/NEJMp1608326 16/x
We found high mortality in both groups: 37 (19%) intv, 33 (15%) control died. 89% of those who died in interv housed vs 29% control. If we target these v high users, may be too late to change mortality, but at least don’t die homeless. We need earlier interventions! 17/x
Take home: PSH works really well even in the users. Need flexibility—housing doesn’t happen overnight & 1st placement may not stick. We shouldn’t expect huge in acute healthcare--& be skeptical. The goal of PSH is to end homelessness 4 ppl w needs, & it succeeds. 18/x
users. Need flexibility—housing doesn’t happen overnight & 1st placement may not stick. We shouldn’t expect huge in acute healthcare--& be skeptical. The goal of PSH is to end homelessness 4 ppl w needs, & it succeeds. 18/x
Will need diff interventions for the 14% of the very highest users (not overall chronic homeless)—who need more intensive/different interventions. But—14% of top 5% is small group. Not only CAN we house those w significant substance use & mental health problems—we must. 19/x
Finally: teamwork makes the dreamwork. All props to @abode_services @sccgov @dstnhome @ChicoineLouis @yecatsmurphy @jenloving23 & many others! We know what to do, we just need to do it /fin
