Evaluation of the Pilot Partnership between HASA and HHC-COBRA. Feasibility Report. prepared by: October 10, 2006

Transcription

1 Evaluaton of the Plot Partnershp between HASA and HHC-COBRA Feasblty Report prepared by: Salle Adams Erc Dovak Terry Hamlton Aqulno Gabor October 0, 2006 Ths plot was made possble by the tremendous support of Else Del Campo (Executve Deputy Commssoner for HASA), Irs Hernandez (Deputy Commssoner for MICSA) and Joanna Om (HHC s Senor Assstant Vce Presdent of Corporate Plannng and HIV Servces). HHC-COBRA Drectors Shemell Castro (North Brooklyn Hosptal Network), Eshelle Tllery (Queens Hosptal Network) and Marlon LeeChong (Metropoltan Hosptal) and HASA Drectors Wllam Mllan (Queensboro Center), Jennfer Carroll (Brownsvlle Center) and Jance Scott (Greenwood Center) provded nvaluable assstance.

2 Evaluaton of the Plot Partnershp between HASA and HHC-COBRA Feasblty Report Executve Summary 4 Introducton 5 Background and Importance 5 Collaboraton between Medcal and Socal Case Managers 5 Goals and Prncpal Fndngs 6 Why Collaboraton Led to Better Outcomes 7 Detals of HASA and HHC-COBRA s Collaboraton 7 Summary of Emprcal Fndngs 9 HIV Prmary Care Appontments 9 Emergency Housng 0 Mental Health Treatment Appontments 2 Substance Abuse Treatment Appontments 4 Concluson and Recommendatons 5 Appendx A: Emprcal Methods and Fndngs 6 Data Sources 6 Model Specfcatons 6 Emprcal Fndngs 7 HIV Prmary Care Appontments 7 Emergency Housng 8 Mental Health Treatment Appontments 8 Substance Abuse Treatment Appontments 9 Appendx B: the Bmodal Logt Model 20 Works Cted 24 2

3 Table descrptons of the varables 25 Table 2 summary statstcs 26 Table a percentage of HIV prmary care appts. kept (bmodal logt model) 27 Table b percentage of HIV prmary care appts. kept (standard logt model) 28 Table 4a clent requred emergency housng (bnary logt model) 29 Table 4b clent requred emergency housng (tests for borough effect ) 0 Table 4c expected number of clents who wll need emergency housng (smulaton) Table 5a percentage of mental health appts. kept (bmodal logt model) 2 Table 5b percentage of mental health appts. kept (standard logt model) Table 6 clent kept all substance abuse treatment appts. (bnary logt model) 4 Table 7 responses to the clent satsfacton survey 5

4 Executve Summary In February 2005, the COBRA programs at the New York Cty Health and Hosptals Corporaton (HHC) began enrollng clents n a plot partnershp wth the HIV/AIDS Servces Admnstraton (HASA) of the New York Cty Human Resources Admnstraton (HRA). Under the framework of the plot, the HHC-COBRA programs at the Queens and North Brooklyn Hosptal Networks collaborated wth HASA to help mutual clents persst n medcal and behavoral health care and to meet the long-term housng needs of mutual clents. Through recprocal tranng on the servces that each organzaton provdes, dstrbuton of admnstratve contact lsts, case conferences and collecton of data on plot partcpants, the plot replaced the nformal bass on whch HASA and HHC-COBRA programs usually work together wth a formal relatonshp. By creatng a formal workng relatonshp between HASA and HHC-COBRA programs, the plot: elmnated duplcaton of effort, kept clents connected to medcal and behavoral health care and helped clents who needed to relocate avod emergency housng. Specfcally, contact lsts and case conferences enhanced communcaton between the organzatons and elmnated duplcaton of effort, whle recprocal tranng helped case management staff collaborate on the cases of mutual clents. Enrollment n the plot ncreased the average clent s probablty of keepng a medcal appontment by about 25 percentage ponts. The plot probably also ncreased the average clent s probablty of keepng a mental health appontment. (Dfferences n the way HHC-COBRA stes defne a clent s need for mental health treatment may have based upward our measurement of the degree of success). The plot was phenomenally successful n reducng a clent s probablty of requrng emergency housng. If the clents studed consttute a representatve sample of HASA clents, expanson of the plot to the entre populaton of HASA clents would cut the ncdence of emergency housng to about half of ts current level. Ths s partcularly mportant because HIV-postve ndvduals who are unstably housed have a hgher probablty of ntravenous drug use and a hgher probablty of tradng sex for money, drugs or housng (Adala et al. Housng Status 2005) and because homeless HIV-postve ndvduals utlze emergency rooms and npatent care more frequently than other HIV-postve ndvduals (Masson et al. 2004). Therefore, by reducng the ncdence of emergency housng, expanson of the plot to other HASA stes and HHC facltes has the potental to slow the rate of HIV transmsson and reduce the ncdence of emergency room vsts and hosptalzatons among HIV-postve ndvduals. Expanson of the plot also has the potental to substantally reduce HASA expendtures on housng. Such a potental arses because plot partcpants who needed to relocate had a lower probablty of requrng emergency housng, whch s more expensve than prvate market housng. Most mportantly, by meetng clents long-term housng needs and by helpng clents adhere to medcal and behavoral health care, expanson of the plot to other HASA centers and HHC facltes has the potental to mprove clents qualty of lfe. 4

5 Introducton Background and Importance HIV-postve patents who receve case management, transportaton, mental health treatment and substance abuse treatment tend to persst n medcal care longer than patents who do not receve such servces (Sherer et al., 2002). Such research suggests that collaboraton between medcal and socal case managers can ncrease the frequency at whch patents keep ther HIV prmary care appontments and therefore help patents acheve better health outcomes. In an effort to create the necessary collaboraton between medcal and socal case management teams, the COBRA programs at the New York Cty Health and Hosptals Corporaton (HHC) and the HIV/AIDS Servces Admnstraton (HASA) of the New York Cty Human Resources Admnstraton (HRA) developed a plot project wth the prmary goal of helpng clents establsh and keep ther medcal appontments. To reach that goal, the plot also amed to help clents avod emergency housng and tred to ensure that clents keep ther mental health and substance abuse treatment appontments. The plot replaced the nformal bass on whch HASA and HHC-based COBRA programs usually work together wth a formal relatonshp. The plot s structure elmnated duplcaton of effort, reduced HASA expendtures on emergency housng and kept clents connected to medcal and behavoral health care. Because the plot reduced the probablty that clents wll requre emergency housng, expanson of the plot to other HASA stes and HHC facltes has the potental to slow the rate of HIV transmsson and reduce the ncdence of emergency room vsts and hosptalzatons among HIV-postve ndvduals. Such a potental arses because HIV-postve ndvduals who are unstably housed have a hgher probablty of ntravenous drug use and a hgher probablty of tradng sex for money, drugs or housng (Adala et al. Housng Status 2005) and because homeless HIV-postve ndvduals utlze emergency rooms and npatent care more frequently than other HIV-postve ndvduals (Masson et al. 2004). Collaboraton between Medcal and Socal Case Managers Prevous research suggests that collaboraton between medcal and socal case managers can mprove the health outcomes of patents lvng wth HIV. When socal case managers ensure that patents have stable housng and ncome and when medcal case managers ensure that patents receve treatment for any mental health and/or substance abuse ssues that they have, ther combned efforts enable patents to meet wth ther physcan more regularly and adhere to ther regmen of medcatons. Patents who adhere more strngently to ant-retrovral therapy tend to have better health outcomes than patents who do not persst n care. Paterson et al. (2000) found that patents who were more adherent to treatment were less lkely to develop HIV nfectons that are resstant to antretrovral drugs. Adherence to medcaton also requres regular consultaton wth a physcan, so a program desgned to mprove the health outcomes of patents must also ensure that patents keep ther HIV prmary care appontments regularly. The degree to whch a patent perssts n care n turn depends on the support servces that he/she receves. Sherer et al. (2002) analyzed clncal data and found that patents who receved case management, transportaton servces, mental health treatment and treatment for chemcal dependency were sgnfcantly more lkely to receve any care, to receve regular care and had more vsts than patents that dd not 5

6 receve those servces. Patents n ther study who receved those servces also had hgher retenton rates than clents who dd not receve those servces. Other research has shown that stable housng and socal support (.e. havng someone to confde n) also play key roles n ncreasng the rate at whch patents adhere to ther regmens of medcatons. Knowlton et al. (2006) studed the lnks between housng, socal support, antretrovral therapy and health outcomes n a sample of njecton drug users and found that socal support plays a major role n facltatng effectve use of recommended hghly actve ant-retrovral therapy (HAART). Of the partcpants on HAART, those who receved strong socal support and stable housng had a much hgher probablty of achevng an undetectable plasma vral load than those who dd not receve strong socal support and stable housng (after controllng for other ndvdual, nterpersonal and structural factors). Knowlton et al. also found that outpatent drug treatment also ncreased a patent s probablty of havng an undetectable plasma vral load, but the effect of drug treatment was not as large as the effects of socal support and stable housng. Goals and Prncpal Fndngs Taken together, the studes cted above suggest that collaboraton between medcal and socal case managers can mprove the health outcomes of patents by ensurng that patents have stable housng and ncome and receve treatment for any mental health and/or substance abuse ssues that they have. The plot project s prmary goal was to mprove the health outcomes of partcpatng clents by ensurng that the clents attend at least 80 percent of ther HIV prmary care appontments. To enable the clents to acheve the desred attendance rate, the plot also sought to ensure that clents keep mental health and substance abuse treatment appontments and ensure that they obtan permanent housng. The plot successfully met these goals. Enrollment n the plot ncreased the average clent s probablty of keepng a medcal appontment by about 25 percentage ponts (a statstcally sgnfcant ncrease). Partcpants n the plot also had a much lower probablty of requrng emergency housng than clents who were not enrolled n the plot. In fact, a smulaton predcts that expanson of the plot to the entre populaton of HASA clents would cut the ncdence of emergency housng to about half of ts current level. Such a predcton assumes that the clents studed are a representatve sample of HASA clents. The plot probably also ncreased a clent s probablty of keepng a mental health appontment, but the measured ncrease may have been based upward by dfferences n the way HHC-COBRA stes defne a clent s need for mental health treatment. Smple averages suggest that there was no statstcally sgnfcant dfference between the rates at whch plot clents and control group clents kept ther substance abuse treatment appontments, but the small sample szes prevented us from makng comparsons whch hold all other factors constant. The collaboraton between HASA and HHC-COBRA enabled plot clents to ncrease the frequency at whch they keep ther HIV prmary care appontments and mental health appontments. Collaboraton also reduced the ncdence of emergency housng among plot clents who needed to relocate. When the plot project was conceved n late 200, ensurng that clents obtaned and retaned Medcad, Publc Assstance and Food Stamps benefts was dentfed as another need. However, a change n HASA recertfcaton procedures greatly mproved beneft retenton and obvated the need to focus on ths ssue. 6

7 Why Collaboraton Led to Better Outcomes In ntervews, HASA and HHC-COBRA staff and admnstrators suggested several reasons why plot partcpants mght acheve better outcomes than clents who are not enrolled n the plot. One explanaton for the plot s success s that the plot mproved the workng relatonshp between HASA and HHC-COBRA staff through ndvdual contacts, recprocal tranng on the servces that each organzaton provdes and dstrbuton of contact lsts (so that case managers could quckly reach the approprate admnstratve staff at the other organzaton). Cooperaton between HASA and HHC-COBRA staff elmnated duplcaton of effort and enabled each organzaton to specalze n provdng ts core set of servces. HASA and HHC-COBRA share the goal of helpng people wth HIV/AIDS and ther famles get the servces they need to reman healthy and ndependent, but they dffer n the servces that they provde. HHC-COBRA offers case management wth supportve servces, such as: prmary medcal care, mental health treatment, substance abuse treatment and counselng. HASA specalzes n ssung welfare benefts such as: Medcad, food stamps, publc assstance and housng. HHC-COBRA provdes assstance wth housng searches, but s not a housng provder. HASA lnks clents to medcal and behavoral health care, but s not a provder of such servces. Consequently, HASA and HHC-COBRA servces complement each other and the ntegraton of HASA and HHC-COBRA teams generates a comprehensve case management servce. Case conferences also helped plot clents acheve better outcomes because clents whose cases were dscussed n case conferences came to the attenton of the HASA Center Drectors, the HHC-COBRA Drectors and all of the case management staff. Durng the conferences, a mutual servce plan was dscussed and case responsbltes were assgned to prevent duplcaton of effort. The ncreased attenton and coordnated servce delvery then led to a better outcome for those clents. Another key to the plot s success was measurement of clent outcomes. Collecton of data from HASA and HHC-COBRA teams helped each team focus on meetng the plot s goals. Detals of HASA and HHC-COBRA s Collaboraton Preparatory Work: The preparaton that occurred pror to enrollment of clents n the plot was one of the keys to the plot s success. One element of the preparatory work was recprocal tranng on the servces provded by each organzaton. Several months pror to the start of the plot, HASA provded a basc one-day orentaton on HASA servces to HHC-COBRA staff and admnstrators ncludng Servcelne s ntake process, elgblty requrements, housng servces, vocatonal rehabltaton servces, emergency housng and Far Hearngs. Of prmary mportance was the tranng and gudance the HHC-COBRA staff was gven regardng the nspecton of apartments. Knowledge of requred documentaton enabled HHC-COBRA case managers to fnd a sutable apartment for plot clents. 7

8 HHC-COBRA admnstratve staff vsted HASA centers to provde half-day tranng to HASA staff on HHC and HHC-COBRA program servces. Conductng the tranng n the HASA centers nvolved n the project ntroduced HHC-COBRA staff to HASA staff and famlarzed them wth the HASA centers. Contact Lst: Dstrbuton of an admnstratve contact lst also enhanced communcaton between HASA and HHC-COBRA staff. The contact lst enabled case managers to easly access nformaton and ndvduals at the other organzaton, prevented losses of tme and helped case managers tell clents about the servces avalable at the other organzaton. The contact lst was essental because penetratng a large organzaton lke HASA can be dffcult and confusng. Pror to the plot, many members of the HHC-COBRA staff ddn t understand the servces HASA provdes and they found t dffcult to reach HASA case managers. Over the course of the plot, both HASA and HHC-COBRA staff found that the contact lst helped them quckly resolve complcated problems because they could access management staff more easly. By February 2005 the preparatory work was complete and COBRA case managers at the Woodhull Medcal and Mental Health Center n Brooklyn and the Elmhurst Hosptal Center n Queens began enrollng HASA clents from the Brownsvlle, Greenwood and Queensboro stes n the plot. Over the course of the plot, a total of 5 clents were enrolled. Case Conferences: Over the course of the plot, formal case conferences were held on a monthly bass so that senor HASA and HHC staff could meet wth HASA and HHC-COBRA case management staff to dscuss some of the more complcated cases and develop a servce plan for those clents. For the less complcated cases, HASA and HHC-COBRA case managers held nformal case conferences over the telephone or durng vsts to a clent s home. Case conferences reduced the problem of duplcaton of effort and enabled HASA and HHC-COBRA staff to focus on provdng ther organzaton s core set of servces. Mult-Dscplnary Case Conferences: On two occasons, the plot convened a mult-dscplnary case conference (MDCC) so that HASA and HHC-COBRA staff and admnstrators could dscuss cases wth the clents prmary care physcans and mental health provders. The MDCCs provded HASA staff and admnstrators wth a unque opportunty to ask questons about ther clents medcal and mental health. Such an opportunty was partcularly valuable because HASA cannot obtan medcal and psychatrc evaluatons performed by hosptal provders unless the clent consents to ther release. Even when HASA obtans the necessary release, t only obtans a wrtten record. By contrast, clents who enroll n HHC-COBRA consent to the release of ther medcal and mental health records at ntake, so HHC-COBRA drectors and case managers can speak drectly wth a clent s prmary care physcan or psychatrst. Contact s further facltated by the fact that clents usually receve ther medcal and mental health care at the same HHC faclty where they receve HHC-COBRA servces. HASA has never had such access to a clent s prmary care physcan or psychatrst, but at the MDCCs, HASA staff and admnstrators could nqure about clents progress n medcal and mental health care. The face-to-face nteracton helped HASA adjust ts servce plan to meet the clents medcal and mental health needs. 8

9 For example, durng a dscusson between a mental health provder and a HASA center drector about one partcular clent s competency to make decsons, the HASA center drector decded to refer the clent to HRA s Offce of Health and Mental Hygene for a psychatrc evaluaton, whch (n ths partcular clent s case) would be used to determne whether or not the clent needs a court-apponted guardan. After the MDCCs, partcpants were asked to provde ther thoughts and opnons about how the MDCC contrbuted to plannng treatment for clents and to explan what they learned from the MDCC. The comments were overwhelmngly postve and tended to stress the dfferent perspectve of the clent that the partcpants heard and the comprehensve nature of the servce plan that was formed at the MDCC. Data Collecton: Fnally, HASA and HHC-COBRA case managers were expected to report on ther clents progress towards meetng the goals of the plot. The accountablty that data reportng provded lent credblty to the project and ensured that HASA and HHC-COBRA delvered on ther commtments to clents by remndng case managers of the outcomes clents were expected to acheve. Summary of Emprcal Fndngs The reports that HASA and HHC-COBRA case managers provded on plot clents and an dentcal set of reports on clents at HHC s Metropoltan Hosptal (whch served as a control group) were combned wth nformaton from HASA s Factors database and Welfare Management System (WMS) database to create the dataset used to evaluate the plot s success n meetng ts goals. The dataset was used to examne the effects that plot partcpaton and other varables had on clents probablty of keepng medcal and behavoral health care appontments and on clents probablty of enterng emergency housng. (Appendx A contans a detaled descrpton of data sources and methodology). It should be noted that the only appontments data that we could obtan reflects the nformaton that clents provded to ther HHC-COBRA case managers. We were unable to obtan more relable data because HHC-COBRA case managers generally do not schedule appontments for ther clents. Data collected by such a method nevtably contans error, but we do not beleve that better data collecton would fundamentally alter the results because the majorty of clents ether kept all of ther appontments or ddn t schedule any at all. Computerzed records would also reflect such a pattern had they been kept. HIV Prmary Care Appontments Goal: Partnershp clents wll keep at least 80% of ther HIV prmary care appontments. By the most conservatve estmate, the average clent s probablty of keepng an HIV prmary care appontment was: 87 percent f he/she was n the plot and 6 percent f he/she was not n the plot. The dfference of 24 percentage ponts s statstcally sgnfcant. The responses of plot clents to a clent satsfacton survey support the fndng that plot clents are more lkely to make and keep medcal appontments. 2 of 42 plot clents (76 percent) ndcated that they began keepng more HIV prmary care appontments snce they enrolled n the plot and 8 of 46 plot 9

10 clents (8 percent) ndcated that ther relatonshp wth ther prmary care provder mproved as a result of the plot. In lght of Sherer et al. s (2002) research (dscussed above), one can attrbute the plot s success n keepng clents connected wth ther prmary care physcans to the collaboratve efforts of HASA and HHC-COBRA to ensure that clent s key needs.e. housng, ncome and medcal nsurance were met. The regresson analyss also ndcates that clents who need substance abuse treatment are less lkely to keep ther HIV prmary care appontments than otherwse dentcal clents who do not have substance abuse ssue. The dfference s statstcally sgnfcant. In dscussons of ths fndng, HASA and HHC admnstrators frequently asked f clents who adhered to substance abuse treatment were more lkely to keep ther HIV prmary care appontments. Unfortunately, our dataset does not have enough chemcally dependent clents to examne the relatonshp between adherence to substance abuse treatment and adherence to HIV prmary care. Sherer et al. studed ths relatonshp and found that HIV-postve patents who needed and receved counselng for chemcal dependency saw HIV prmary care physcans sgnfcantly more often than patents who needed but dd not receve counselng. Patents who needed and receved counselng were ntally more lkely to receve regular medcal care, but were less lkely to receve regular medcal care n the second year of ther study perod (as compared to patents who needed but dd not receve counselng). Emergency Housng Goal: HASA and HHC-COBRA Case Managers wll form a plan for permanent housng and collaborate to obtan permanent housng wthn days of the clent s readness and avalablty of permanent housng. Accordng to Adala et al. (CHAIN Update Report #4, 200), HIV-postve ndvduals wth a hstory of housng needs who receve housng assstance are much more lkely to obtan medcal care and persst n care than those who do not get housng assstance. Such a fndng helps explan why Knowlton et al. (2006) found that ndvduals wth stable housng had lower vral loads than those who dd not. Such research ndcates that placement of patents n stable housng supports the plot s goals of keepng clents connected to medcal care and of helpng them lead healther lves. 0

11 Stable housng s also less expensve than emergency housng. Accordng to HASA admnstrators, HASA pays a commercal hotel an average of $620 per month to house a sngle clent on an emergency bass. For comparson, housng a sngle clent n an unsubsdzed prvate market apartment only costs an average of $07 per month. In addton to reducng the cost burden mposed on HASA by a hgh ncdence of emergency housng, placng clents n stable housng also has the potental to reduce the rate of transmsson of HIV because HIV-postve ndvduals who are unstably housed are more lkely to use ntravenous drugs and engage n prosttuton (Adala et al. Housng Status 2005). Because homeless HIV-postve ndvduals vst emergency rooms and requre hosptalzatons more frequently than those wth some form of housng (Masson et al. 2004), placng clents n stable housng also has the potental to reduce the costs assocated wth provdng acute care to homeless HIV-postve ndvduals. Because only 5 clents n our dataset requred emergency housng at any pont n tme, the sample sze was too small to evaluate the plot s success n movng clents out of emergency housng. It was however feasble to examne the plot s success n preventng clents from requrng emergency housng. The plot was tremendously successful on ths measure. By the most conservatve estmate, the probablty that the average clent (who needs to move) wll requre emergency housng was: 4 percent f he/she was n the plot and 68 percent f he/she was not n the plot. The dfference of 4 percentage ponts s statstcally sgnfcant. To estmate the mpact that replcaton of the plot at all HASA stes would have on the ncdence of emergency housng, we assumed that all of the clents n the dataset need to move and computed the expected number of clents who would need emergency housng under two scenaros: one n whch all of the clents are enrolled n the plot and one n whch none of the clents are enrolled n the plot. The smulaton predcts that f the clents n our dataset are a representatve sample of HASA clents then replcaton of the plot at all HASA stes would cut the need for emergency housng n half. More specfcally, the scenaro n whch all clents are enrolled n the plot yelds an expected number of clents

12 who would need emergency housng that s half as large as the expected number obtaned from the scenaro n whch none of the clents are enrolled. The assumpton that clents n our dataset are a representatve sample of HASA clents should not be understated. For example, the regresson models that we estmated also ndcate that clents who do not speak Englsh well are less lkely to requre emergency housng than clents who speak Englsh fluently. Therefore, replcaton of the plot n predomnantly Spansh-speakng neghborhoods wll reduce the need for emergency housng n those neghborhoods, but the reductons n those neghborhoods wll be smaller than reductons n Englsh-speakng neghborhoods. The plot s success n preventng clents from requrng emergency housng can be attrbuted to both the tranng that HHC-COBRA staff receved on HASA housng gudelnes and to the sprt of cooperaton that the plot helped to foster. Under the framework of the plot, HHC-COBRA case managers are responsble for assstng wth housng searches and work wth HASA case managers to develop a plan to place clents n permanent housng. Consequently, the plot streamlned the assstance a clent receves n fndng a new place of resdence. Responses to the clent satsfacton survey also shed lght on the ways n whch the plot helped them obtan permanent housng. Of the 27 respondents to a queston on referrals to permanent housng, 5 plot clents (56 percent) ndcated that they receved a referral from HASA and clents (48 percent) ndcated that they receved a referral from ther HHC-COBRA program. Only three clents ( percent) sad that they dd not receve a referral from ether HASA or HHC-COBRA. Of the 28 respondents to a queston on housng assstance, 8 clents (64 percent) ndcated that HASA gave them a lot of housng assstance and 2 clents (82 percent) ndcated that HHC-COBRA gave them a lot of housng assstance. Mental Health Treatment Appontments Goal: Partnershp clents wll keep at least 70% of ther behavoral health treatment appontments where applcable. Paterson et al. (2000) found that mental llness reduced the rate at whch patents adhere to protease nhbtor therapy. Sherer et al. (2002) found that patents whose need for mental health care was met were more lkely to receve regular medcal care than patents whose need for mental health care went unaddressed. Such research ndcates that provdng mental health care (when approprate) keeps clents wth mental llness connected to ther HIV prmary care physcans. The plot seems to have sgnfcantly ncreased a clent s probablty of keepng mental health appontments, but not to the 70 percent level. By the most conservatve estmate, the average clent s probablty of keepng a mental health treatment appontment was: 56 percent f he/she was n the plot and 4 percent f he/she was not n the plot. The dfference of 52 percentage ponts s statstcally sgnfcant. 2

13 Although the goal of 70 percent was not met, plot clents were substantally more adherent to mental health treatment than non-plot clents. The large dfference can be attrbuted n part to the collaboraton between HASA and HHC-COBRA. Of the 9 respondents to a clent satsfacton survey queston on mental health care, clents (58 percent) sad that they began keepng more mental health treatment appontments snce they enrolled n the plot. However, part of the dfference may be attrbutable to dfferences n the crtera that HHC-COBRA stes use to determne whch clents should be referred to mental health treatment. To see how dfferences n crtera may have affected our estmate of the average clent s probablty of keepng a mental health appontment, magne that the HHC-COBRA case managers at Metropoltan Hosptal (the ste of the control group) referred all clents who have borderlne mental llness to treatment, whle HHC-COBRA case managers at the North Brooklyn and Queens Hosptal Networks (the plot program stes) ddn t refer any clents who have borderlne mental llness to treatment. Imagne further that all clents who have borderlne cases of mental llness refuse treatment (.e. they keep zero percent of appontments), whle clents who have more severe cases of mental llness keep all of ther appontments. In such a scenaro, the efforts of HHC-COBRA case managers at Metropoltan Hosptal to place clents n mental health treatment would have reduced the control group s average percentage of appontments kept. Such an extreme scenaro s unlkely to have occurred, but t llustrates the way n whch the defnton of need for mental health treatment can affect the measurement of a clent s predcted probablty of keepng a mental health treatment appontment. The regresson analyss also ndcates that clents who need help managng ther fnances and clents who need treatment for substance abuse have a lower probablty of keepng mental health appontments. Once agan, the small number of chemcally dependent clents n our dataset prevents us from examnng the relatonshp between adherence to substance abuse treatment and adherence to mental health treatment. Fnally, the regresson analyss ndcates that motherhood lowers a clent s probablty of keepng a mental health treatment appontment and that lvng wth another adult ncreases a clent s probablty of keepng an appontment. However the two effects do not cancel out. Lvng wth another adult ncreases a

14 mother s probablty of keepng a mental health treatment appontment, but not to the level that would preval f she were not a mother. Substance Abuse Treatment Appontments Goal: Partnershp clents wll keep at least 70% of ther behavoral health treatment appontments where applcable. In a sample of 85 former and current drug users, Arnsten et al. (2002) found that HIV-postve ndvduals who cope wth stress by consumng alcohol and llegal drugs tended to be less adherent to hghly actve ant-retrovral therapy (HAART) and had hgher vral loads. In partcular, they found that actve cocane use was strongest predctor of poor adherence. Actve users of heron were also less adherent to therapy, but the dfference was not statstcally sgnfcant. Other studes have not been able to draw a frm lnk between substance abuse treatment and adherence to ant-retrovral therapes however. Moatt et al. (2000) found that njecton drug users on buprenorphne drug mantenance treatment were more adherent to HAART than former njecton drug users. The authors cauton however that physcans who treated the sample s patents were very reluctant to prescrbe HAART to current njecton drug users and may only have prescrbed t to the ones who were lkely to be adherent to both HAART and drug mantenance. Sherer et al. (2002) was smlarly unable to draw a frm lnk between substance abuse treatment and retenton n medcal care. They found that chemcally dependent patents who receved counselng had a hgher number of total vsts to ther HIV prmary care physcans than those who dd not receve counselng, but were less lkely to receve regular care. Despte the lack of frm lnks, one cannot dsmss the possblty that addressng ssues of chemcal dependency wll help clents adhere to ant-retrovral therapy and persst n care. Unfortunately, our dataset doesn t shed any lght on the ssue. As mentoned prevously, the number of clents n our dataset who need substance abuse treatment s too low to examne the relatonshp between adherence to substance abuse treatment and persstence n medcal care. Two dffcultes hampered our ablty to examne whether or not partcpaton n the plot ncreased clents probablty of keepng substance abuse treatment appontments. Frst, although plot clents had a hgher average rate of adherence to treatment than control group clents, the dfference s not statstcally sgnfcant because the number of chemcally dependent clents n the dataset s too small. Second, of the 5 clents who need substance abuse treatment, 49 ether kept all of ther appontments or none at all. The ones who kept all of ther appontments were generally n methadone mantenance. Respondents to the clent satsfacton survey dd however ndcate that the plot helped them to adhere to substance abuse treatment. Of the 5 respondents, 0 clents (67 percent) sad that they began keepng more substance abuse treatment appontments snce they enrolled n the plot. 4

15 Concluson and Recommendatons By creatng a formal workng relatonshp between HASA and HHC-COBRA, the plot program fostered a sprt of cooperaton among the case management staff. Contact lsts and case conferences enhanced communcaton between the organzatons and elmnated duplcaton of effort, whle recprocal tranng on the servces that each organzaton provdes helped case management staff collaborate on the cases of mutual clents. Collaboraton between HASA and HHC-COBRA ncreased plot partcpants probablty of keepng a medcal appontment. The plot probably also ncreased the probablty that a clent wll keep a mental health appontment (although defntonal ssues cloud the degree of success). Consequently, expanson of the plot to other HASA centers and HHC facltes has the potental to: help clents reman connected to ther HIV prmary care physcans and help clents reman connected to ther mental health care provders. Because the plot reduced the probablty the ncdence of emergency housng among partcpants who needed to move, expanson of the plot has the potental to: substantally reduce HASA expendtures on emergency housng, help clents avod emergency room vsts and hosptalzatons and reduce the rate of HIV transmsson. Although there were not enough clents n our dataset to evaluate the plot s success n movng clents out of emergency housng, the plot s success n helpng clents avod emergency housng and the tranng that HHC-COBRA staff receved on HASA housng gudelnes ndcate that the partcpaton n the plot has the potental to meet the long-term housng needs of clents resdng n emergency housng. Clents who resde n emergency housng should therefore be encouraged to enroll durng the next phase of the plot. Enrollment should help them obtan long-term medcally-approprate housng, enable them to persst n medcal and behavoral health care and most mportantly mprove ther qualty of lfe. 5

16 Appendx A: Emprcal Methods and Fndngs Data Sources Data on partcpants n the plot (who were enrolled n HHC-COBRA programs at the North Brooklyn and Queens Hosptal Networks) and data on a control group of clents (who were enrolled n the HHC- COBRA program at Metropoltan Hosptal) was taken from several sources. The most mportant source of data was a short form that HASA and HHC-COBRA case managers completed. The short forms provded us wth basc demographcs, language, medcal statstcs (e.g. vral loads and CD4 counts), nformaton about the clents lvng stuaton, nformaton about whether the clent has needs substance abuse and/or mental health treatment, the number of appontments a clent made and kept and an assessment of the clent s ablty to perform the actvtes of daly lvng. HHC-COBRA case managers generally do not schedule appontments for ther clents, so the case managers had to obtan appontments data by askng ther clents how many appontments they made and kept. Clents who dd not schedule any appontments at all were assumed to keep zero percent of ther appontments. Another mportant source of data was HASA s Factors database. Factors provded us wth nformaton on clents housng status and when the date clents were dagnosed as beng HIV-postve symptomatc. Fnally, HHC-COBRA records provded us the dates when clents entered the plot. Table provdes descrptons of the varables created from these data sources. Model Specfcatons For gudance n selectng the varables used n the regresson models, we turned to prevous research. Moatt et al. (2000) found that younger clents, clents who consumed alcohol and clents who had negatve lfe events n the prevous sx months tended to be less adherent to antretrovral therapy. Sherer et al. (2002) found that less regular care occurred more frequently among women, younger patents and ntravenous drug users. On the bass of these studes and the data avalable to us, we decded to nclude age, gender and substance abuse ssue n each of our model specfcatons. We also chose to control for whether or not the clent speaks Englsh well and the degree to whch a clent needs assstance n managng hs/her fnances. We also hypotheszed that women may attend care less frequently f they are sngle mothers, so an alternatve specfcaton replaces gender wth a varable whch ndcates whether or not a clent s a mother. Wth one excepton, a varable that ndcates whether or not the clent lves wth another adult was ncluded n all model specfcatons because the other adult may provde assstance n carng for chldren and may also provde moral support and encouragement to the clent. (We had to exclude the varable that ndcates whether or not a clent lves wth another adult from our regresson on a clent s probablty of keepng a substance abuse treatment appontment to ncrease the number of ncluded observatons). 6

17 We also observed a postve correlaton between the need for substance abuse treatment and the need for mental health treatment (the smple correlaton coeffcent for these two dummy varables s 4 percent). Of the 5 clents who need substance abuse treatment, 4 also need mental health treatment. To avod ntroducng near sngularty nto the covarance matrx and to dscern whether t s substance abuse or mental llness whch affects a clent s probablty of keepng an appontment or need for emergency housng, we decded not to nclude both substance abuse treatment needs and mental health treatment needs n the same regresson specfcaton. Fnally, many clents kept all of ther HIV prmary care, mental health appontments and substance abuse treatment appontments. Many others dd not schedule any appontments at all. Consequently, the modes occur at 0 and 00 percent, but there s also a relatvely large number of clents who kept some, but not all, of ther appontments. Snce standard bnary choce models assume a symmetrc unmodal densty functon (.e. one that predcts that the majorty of observatons wll le close to ther mean), we had to use a dstrbuton that yelds a bmodal densty functon, because a bmodal densty functon predcts more observatons far from the mean than observatons close to the mean (Appendx B descrbes the specfc dstrbuton that we used). Emprcal Fndngs HIV Prmary Care Appontments Accordng to the second column of regresson results n Table a, the average clent has an 87. percent probablty of keepng an HIV prmary care appontment f he/she s n the plot and a 6. percent probablty f he/she s a clent n the control group at the Metropoltan Hosptal. The 2.9 percentage pont dfference (standard error: 8. percentage ponts) s the smallest predcted dfference. Dfferent specfcatons suggest that the plot was slghtly more successful. The estmated coeffcents (n the upper panel) that have stars next to them are the varables that have a statstcally sgnfcant effect on the dependent varable (n ths case: the frequency at whch a clent kept HIV prmary care appontments). Statstcal sgnfcance essentally means that there s a hgh degree of certanty that the coeffcent s not zero. In ths case, the only two statstcally sgnfcant varables are Met clent and needs SA treatment. The fact that the estmated coeffcent of Met clent s negatve means that clents at Metropoltan Hosptal (.e. clents n the control group) have a lower probablty of keepng HIV prmary care appontments than otherwse dentcal clents n the plot. Smlarly, the fact that the estmated coeffcent of needs SA treatment s negatve means that clents who have a substance abuse ssue have a lower probablty of keepng HIV prmary care appontments than otherwse dentcal clents who do not have substance abuse ssue. The F-statstcs n Table b ndcate that standard logt models (whch assume a symmetrc unmodal densty functon) do not explan a sgnfcant porton of the varaton n the percentage of appontments kept (expressed as the log of an odds rato) because so many clents kept ether all or none of ther HIV prmary care appontments. 7

18 Emergency Housng As dscussed n the body of ths report, the plot s record n keepng clents out of emergency housng was phenomenal. By the most conservatve estmate (n the rghtmost column of Table 4a), the average clent who needed to move had a 67.6 percent probablty of requrng emergency housng f he/she was not n the plot and a 4.0 percent probablty of requrng emergency housng f he/she was n the plot (a dfference of.5 percentage ponts, wth a standard error of.9 percentage ponts). To estmate the reducton n the number of clents who need emergency housng, we assumed that all clents n the dataset need to relocate and used the estmated coeffcents (n Table 4a) to compute each clent s probablty of requrng emergency housng under a scenaro n whch each clent s enrolled n the plot and under a scenaro n whch no clent s enrolled n the plot. Because each observaton s ndependent (one clent s housng status does not affect another s), summaton of the probabltes yelds the expected number of clents who wll need emergency housng. The results of the smulaton (lsted n Table 4c) show that the scenaro n whch all clents are enrolled n the plot has half the expected number of clents who would need emergency housng as the scenaro n whch none of the clents are enrolled. Because the clents enrolled n the plot lve n Brooklyn and Queens, whle the clents n the control group lve n Upper Manhattan, we also checked to make sure that the regressons dd not pck up a borough effect. In other words, we checked to make sure that clents n the control group were not more lkely to end up n emergency housng smply because they lve n the relatvely more expensve Upper Manhattan locaton. To perform such a check, we ncluded the Met clent varable n the regresson (Table 4b). The clent n plot before move varable controls for whether or not the clent was enrolled n the plot for at least one month before he/she moved, whle the Met clent varable controls for a clent s borough of resdence. Smple t-tests ndcate that the coeffcent on clent n plot before move s negatve and statstcally sgnfcant, whle the Met clent coeffcent s not statstcally sgnfcant. Furthermore, the lkelhood rato test statstcs also ndcate that the Met clent varable should be excluded from the model whle the clent n plot before move varable should be ncluded. On the bass of such tests, we can conclude that t s enrollment n the plot that reduces a clent s probablty of requrng emergency housng and not a borough effect. Fnally, the coeffcent on the non-englsh was also negatve and statstcally sgnfcant, whch ndcates that clents who do not speak Englsh well have a lower probablty of requrng emergency housng than clents who do speak Englsh well. Mental Health Treatment Appontments Accordng to the rghtmost column of Table 5a, the average plot clent had a 55.7 percent probablty of keepng a mental health appontment, whch substantally hgher than the 4. percent predcted probablty for an dentcal clent at Metropoltan Hosptal a dfference of 5.6 percentage ponts (standard error: 5.6 percentage ponts). The alternatve model specfcaton predcts slghtly more success. 8

19 Although the predcted probabltes of 4. and 55.7 percent are consstent wth the respectve averages of 9.6 and 5.0 percent (see Table 2), there s reason to be skeptcal about ths result. Whle the plot may have had a postve effect on a clent s probablty of keepng mental health appontments, somethng other than the plot seems to be nfluencng the predcted probabltes. For example, consder a clent who has borderlne mental llness who refuses treatment (despte a case manager s referral). Snce the clent has borderlne mental llness, t s by no means clear whether that clent should be classfed as needng treatment or not. If the staff at Metropoltan reported more clents wth borderlne mental llness than the staff at North Brooklyn and Queens, then Metropoltan would have a larger share of refusals and Metropoltan s average percentage of appontments kept would be lower. Examnng the other varables n the regresson, t s nterestng to note that the coeffcent on mother s negatve but larger n absolute value than the coeffcent on lves wth adult. Ths suggests that mothers have a lower probablty of keepng mental health appontments. Lvng wth another adult ncreases a mother s probablty of keepng an appontment, but not enough to overcome the effect of beng a mother (the sum of the two coeffcents s equal to and the standard error of that sum s 0.478). Fnally, one can also see that clents who need treatment for substance abuse and clents who need help managng ther fnances have a lower probablty of keepng mental health appontments. Although the regresson results n Table 5b ndcate that standard logt models explan a sgnfcant porton of the varance of the percentage of appontments kept (expressed as the log of an odds rato), the estmates should be based on a dstrbuton whch predcts that there wll be more observatons far from the mean than observatons close to the mean because the majorty of clents ether kept all of ther appontments or refused treatment (.e. whch s equvalent to keepng zero appontments). The advantage of the bmodal logt model can be seen by comparng R-squared statstcs. The bmodal logt model explans about half of the varaton n the dependent varable, whle the standard logt model only explans about a quarter of the varaton. Substance Abuse Treatment Appontments Of the 2 plot clents who need substance abuse treatment (and for whom we have data), 4 kept all of ther appontments (.e. 44 percent). Of the 7 clents n the control group at Metropoltan Hosptal who need substance abuse treatment, 5 kept ther appontment (.e. 29 percent). The 4 percentage pont dfference s not statstcally sgnfcant from zero because the standard error of the dfference s also 4 percentage ponts. Attempts to use regresson analyss to control for other factors that may have contrbuted to the dfference n the percentage of clents who kept all of ther appontments were unsuccessful. The low lkelhood rato statstcs of the estmated models (n Table 6) ndcate that we cannot reject the hypothess that none of the varables has an effect on a clent s probablty of keepng a substance abuse treatment appontment. There are two reasons for the lack of statstcal sgnfcance. One s the small sample sze. The sample only has 5 clents who need substance abuse treatment. The other reason s methadone mantenance treatment. Of the few observatons that we do have, most of the clents n the sample don t keep any appontments at all and those who do keep appontments are usually gong for methadone mantenance. 9

20 Appendx B: the Bmodal Logt Model As mentoned n the text of ths report, many clents kept all of ther HIV prmary care and mental health treatment appontments, whle many others dd not schedule any appontments at all. There s also a large number of clents that kept some, but not all, of ther appontments. The bmodalty n the appontments seres may be a form of state dependence, whch would arse f a clent s decson to keep an appontment depends on whether or not he/she kept a prevous appontment. Unfortunately, the avalable data does not allow us to estmate an ntertemporal bnary choce model because we only have nformaton on the percentage of appontments made and kept. (Case managers often schedule a clent s ntal appontment, but generally do not schedule follow-up appontments). Snce the largest numbers of observatons occur at 0 and 00 percent, we need to assume that each clent s true probablty of keepng an appontment s at least 5 percent and at most 95 percent. We also need a probablty densty functon (pdf) whch predcts that there wll be more observatons far from the mean than observatons close to the mean. Fortunately, a slght modfcaton of the logstc dstrbuton yelds a vable pdf. The cumulatve dstrbuton Prob ( Y = ) has the bmodal densty functon: λ exp = exp ( β' x) ( β'x) γβ'x) ( β'x) γ ( β'x) ( β'x) ( β'x) = ( β'x) where : 0 < γ < ( ( β'x) γ) ( β'x) ( ( β'x) ) d = d Just as weghted least squares can be used n standard logt models, weghted least squares can also be used n the bmodal logt model. The weghts n the bmodal logt model dffer from those n the standard logt model however. 20

21 Accordng to basc statstcs, the expected value of the observed probablty, P, s equal to the true probablty, π. The expected value of the error term, ε, therefore s zero. The varance of the error term however depends on the probablty tself and the number of observatons, n. Specfcally: P = π ε where : [ ] = 0 and Var[ ε ] E ε = π ( π ) n Consequently, the error terms from an ordnary least squares regresson wll not have constant varance (.e. heteroscedascty wll be present). Greene (2000, p. 84-6) suggests a weghted least squares framework that we can use to correct for the heteroscedascty of the error terms. He uses the fact that the cumulatve dstrbuton (n our case: ( β'x) ) has an nverse (because t s a monotoncally ncreasng functon of β' x ) to obtan the expected value and varance of the error terms. The nverse of ( β' x ) = π s wrtten as ( ) = β'x π. By the nverse functon rule: d d π ( π ) = d π d ( π ) d ( β' x ) d ( β'x ) λ( β' x ) = Greene s framework calls for us to take a Taylor seres approxmaton to the functon ( P ) around the pont where P = π to obtan the regresson equaton: ( P ) ( π ) ε d d π 2 ( π ) ( P π ) ( P ) β' x where : λ λ( β' ) x λ