2017 SIZE ESTIMATION OF HIV KEY POPULATION IN INDONESIA 2016

  F O R W A R D

  exposure (KP) are essential to advocate for resources, prioritize prevention and other programmatic activities among groups, predict the number of persons living with HIV (PLHIV) and the impact of the HIV epidemic through modeling, and assess program coverage. Population size estimates (PSEs) are an important component of surveillance as well as monitoring and evaluation.

  Ministry of Health has started the population size estimation (PSE) since 2002. The PSE methods were kept updated and being improved for obtaining better result. Report of Size Estimates HIV of Key Population in 2016 is a renewal of the report of Size Estimation of Key Affected Population in 2012 issued by the Ministry of Health in year of 2014. This report describes a comprehensive situation and can understand in relation to the size of key affected population up to district level.

  The estimation results in 2012 showed that there were 7.4 to 10.2 million people with a median value around 8.8 million key population. The results of these estimates were then included in the calculation of estimation and projections of HIV/AIDS in Indonesia in 2011-2016. The 2016 PSE exercise was resulting final population size estimate of FSW, MSM, Waria, PWID, and Clients of FSW and Waria which the highest size estimate is clients of FSW which account for 5 million, and the smallest size estimates is PWID which account only 33 thousand. The results of these estimates are also then included in the calculation of estimation and projections of HIV/AIDS in Indonesia in 2015-2020.

  The process of size estimation of the key population has gone through a long and complex process involving a variety of associated partners. This estimation methodology and results have been reviewed by a group of experts and presented to stakeholders. The results of the reviews stated that with all the limitations that exist in the calculation of this estimate, the result is the best result that can be obtained with the data available at the time the calculation is done.

  We express our deepest appreciation to all parties for the attention, assistance and contribution in the preparation, implementation, and improvement of the estimation activities. Hopefully this book useful for the HIV-AIDS control program, not only for the Ministry of Health, but also to all work partners of HIV AIDS control.

  Jakarta, February 2017 Director General Disease Control and Prevention

  dr. H. Mohamad Subuh, MPPM

  NIP 196201191989021001

CONTRIBUTOR LIST

  Anak Agung Sagung Sawitri - WHO/Consultant; Ari Wulan Sari – Ministry of Health; Dwi Rahmadini- National AIDS Comission; Irma Siahaan –National AIDS Comission; Lely Wahyuniar –UNAIDS; Leonita Agustine—UNAIDS; Muhardi Kahar –National Statistic Office; Rizky Hasby –Ministry of Health; Sigit Ari Saputro – Airlangga University; Yori Novrianto –FHI 360.

  WRITER

  Anak Agung Sagung Sawitri– WHO/Consultant

  REVIEWERS

  Siti Nadia Tarmizi – Ministry of Health; Endang Budi Hastuti – Ministry of Health; Triya Novita Dinihari – Ministry of Health; Irawati Panca – Ministry of Health;Gerald Jacobson – WHO/Consultant; Lisa G. Johnston – FHI360/Consultant; Dongbao Yu – WHO; Taoufik Bakkali – UNAIDS; Fetty Wijayanti – WHO; Geoffrey Nan Li – UNICEF; Caroline Francis – FHI 360; Siti Sulami – FHI 360; Indang Trihandini – University of Indonesia; Mondastri Korib – University of Indonesia; Rossy – PHO West Java; Aan Hermawan – PHO West Java.

  CONTRIBUTORS

  Asep Hardiansyah – GWL Ina; Ayu Oktariani – IPPI; Christian – Gaya Dewata; Christian P. – Kios Atmajaya; Citra Wahyuningsih – VCT Sanglah; Cok Istri Sri Dharma Astiti

  • – PHO Bali; D W Wirma – PAC Bali; Dewa Nyoman Wirawan – Udayana / Yayasan KertiPraja; Dewi Dian – VCT Sanglah; Eva Kartikasari – YKS; Fais Abdillah –Yakeba; Fonny J Silvanus – NAC; Gde Agus Suryadinata – PHO Bali; Hartini – IPPI; I Made Adi Wiguna – PHO Bali; I Made Ari Sudana – PAC Bali; I N Kenyem Subagja – PHO Bali; I N Sudiyasa – PHO Bali; Irawati –MoH; Komang Ayu Trisna – VCT Sanglah; Kon Sriwiyartini – RS Sanglah; Luh Putu Musnitarini – PHO Bali; M. Syamsoel – Orbit; Made Ratri – VCT Sanglah; Marstyo Catur – YSS; Ni Luh KD Sri Sastradewi – PHO Bali; Ni Nyoman Sri Sutarmi – VCT Sanglah; Ni PutuPutri Pratiwi – DAC Badung; Novita E Wuntu – Citra Usada Indonesia; Otto Sugiarto – Yayasan DuaHati; Pipiet Laksmono – GWL Ina; Priscillia Anastasia – WHO DKI Jakarta; Puji Suryantini - WHO Indonesia; Rediscoveri – YKB; Resti –Bandungwangi; Retno Mulyaningsih – GF AIDS; Rosidin

  MA– YIM; Saban – Karisma; Saiman – Bandungwangi; Tetty Rahmawati – USAID; Tiara Nisa – WHO Indonesia; Titik Suwarti – MoH; Vinny Sutriani – MoH; Yahya Anshori – PAC Bali; Yenny Tju.

  ExECUTIvE SUmmARy

  HIV prevention program, and also for further projection of people living with HIV (PLHIV) in Indonesia. The 2016 PSE was collaborative works of Ministry of Health (MOH), National AIDS Commission (NAC), National Statistic Office, Universities, within the technical support from WHO, UNAIDS, and FHI Linkages. The results have assessed for face validity by relevant key stake holders and the respective communities.

  Indonesia has started PSE since 2002. The PSE methods were kept updated and being improved for obtaining better result. In 2016, the PSE is adopted the 2012 PSE methods with several improvements for additional vary and updated data sources, implementing theoretical concepts in predictors selection, and implementing adjustment factors. KPs being estimated is included female sex workers (FSW), men have sex with men (MSM), waria, and people who inject drug (PWID). Client of FSW and client of waria were also estimated.

  The 2016 PSE used multivariable regression models to extrapolate estimates based on mapping conducted at locations (“venues”) frequented by KPs in selected districts to other districts where mapping had not been conducted, to arrive at a national PSE. Further adjustments were implemented to the estimates to account for KP members who may not have been reflected in the mapping figures.

  The main data to be used for predicting the size estimates was mapping data and 2015

  IBBS. Thus, the 2016 PSE was starting with assessment of data and the process of obtaining those data. Then, the predictor variables for the extrapolation models were selected based on theoretical framework and statistical significance. This was followed by developing and fitting the regression models to the mapping data and using the fitted models to extrapolate the mapping counts to the unmapped districts. Further adjustments were conducted to account for KP members who do not attend venues, and additional calculations to convert mapping counts to annualized figures that represent the number of KP during a 12-month period. The size estimates result was also assessing for face validity, and revision may also raise from consensus among stake holders and expert.

  The main results of 2016 PSE were as follows: Size estimates of KP obtained from regression model and adjustment; including: • FSW is 226,791 (128,114, 364,313), MSM is 754,310 (648,641, 866,840), waria is 38,928 (13,038, 89,640), PWID is 33,492 (14,016, 88,812).

  Size estimates of KP obtained from multiplying with common values of KP in Asia • and consensus: clients of FSW 5,254,065 (4,415,776, 6,159,431)

  • Size estimates of KP obtained from multiplying PSE result with proportion of KP from IBBS: clients of waria 350,119 (327,596, 375,236)
followings are the main points of discussion: FSW size estimates are stable and similar with the 2012 PSE, and also in line with • the common range size of FSW in Asia MSM size estimate s are far lower than the 2012 PSE, and also lower than the • common range size of MSM in Asia. This is due to the use of consensus inflation number for MSM because no supporting data for reasonable inflation factor currently. In addition, there were arguable of the 2012 PSE inflation factor for MSM.

  Waria size estimates are slightly higher than the 2012 PSE and far lower than other • country. However, this number was acceptable for face validity. PWID size estimates are far lower than the 2012 PSE and other report. However, • there are supporting data to convince the accuracy of this number Clients of FSWs were lower than the 2012 PSE; but still in the common range of • Asia. Most of available data are coming from high risk men, thus consensus should be used to obtain the final estimates Clients of waria were lower than the 2012 PSE; but no supporting data available. •

  The 2016 PSE is not without limitations. Although the statistical methods are sophisticated, the data being used have their originated weaknesses; and these must have influenced the result of the PSE. Likewise, the size estimates produce less reliable in the district level. In addition, there were still many consensuses should be taken during the exercise; which may subject to error.

  To some extent, these are the best numbers could be used in the program planning and evaluation. Considering that these number are still estimates, the understanding how to use the PSE result is needed. Therefore, it is very important to develop the capacity of health staff in provincial and district level in term of PSE and its attributes. Further improvement for main data sources is needed.

  DHS : Demographic and Health Survey DFSW : Direct Female Sex Worker

  IBBS : Integrated Biologic Behavior Survey FHI Linkages : Family Health International Linkage (FHI360) FSW : Female Sex Workers HIV : Human Immunodeficiency Virus

  IDFSW : Indirect Female Sex Worker

  IF : Inflation Factor KP : Key Population MSM : Men Who Have Sex with Men MOH : Ministry of Health NAC : National AIDS Commission NSO : National Statistic Office PODES : Survei Potensi Desa (Survey of Village Potency) PLHIV : People Living with HIV PSE : Population Size Estimate PWID : People Who Inject Drug UNAIDS : United Nations on AIDS WHO : World Health Organization SAKERNAS : Survey Angkatan Kerja Nasional (National Survey of Work

  Labor) SUSENAS : Survei Sosial Ekonomi Nasional (National Survey of Economy and Social)

  GLOSSARy

  TABLE OF CONTENTS

FORWARD .................................................................................................................. iii

CONTRIBUTOR LIST ......................................................................................................... iv

EXECUTIVE SUMMARY .................................................................................................... v

GLOSSARY .................................................................................................................. vii

TABLE OF CONTENTS...................................................................................................... ix

  

1. BACKGROUND ....................................................................................................... 1

  

2. METHODOLOGY ..................................................................................................... 3

  2.1. Objectives ......................................................................................................... 3

  2.2 General Overview of the PSE Process ............................................................. 3

  2.3. Definition of KPs ............................................................................................... 4

  

3. STEPS IN THE CALCULATION OF KPs ............................................................... 7

  3.1. Review of Mapping Data ................................................................................. 7

  3.2. Extrapolation of Mapped Counts to Unmapped Districts Using Regression Models .............................................................................................................. 7

  3.3. Using Inflation Factors to Account for Non-Venue-Going KP .......................... 14

  3.4. Estimating Number of Clients of Female Sex Workers ................................... 17

  3.5. Estimating Number of Clients of Waria ............................................................ 18

  

4. RESULT .................................................................................................................. 19

  

5. DISCUSSION ........................................................................................................... 20

  Annex 1. Mapping and IBBS Sources and Limitations ........................................... 25 Annex 2. Comparison of 2012 with 2016 PSEs. ..................................................... 30 Annex 3. Adjusting district-level KP Size Estimates for Use in Setting Program

  Coverage Targets and Assessing Program Performance ........................ 32 Annex 4. Calculation of Clients Using 2012 PSE Method ...................................... 33 Annex 5. PSE Result by Province ........................................................................... 34 Annex 6. Akaike Information Criterion ..................................................................... 40

1 BACKGROUND

  exposure (KP) are essential to advocate for resources, prioritize prevention and other programmatic activities among groups, predict the number of persons living with HIV (PLHIV) and the impact of the HIV epidemic through modeling, and assess program coverage. Population size estimates (PSEs) are an important component of surveillance as well as monitoring and evaluation.

  In Indonesia, the Sub-Directorate of AIDS and Sexually Transmitted Infections (STIs) of the Ministry of Health (MOH) has conducted several rounds of size estimation, in 2002, 2004, 2006, 2009, and 2012. Indonesia develops size estimates for six populations: men who have sex with men (MSM); female sex workers (FSWs); clients of female sex workers; people who inject drugs (PWID); and, Waria (male-to-female transgender); and clients of waria sex workers. Worldwide, Indonesia is one of 38 countries to produce size estimates at the national level using a well-documented methodology for any KP.

  A key challenge faced by all countries is to develop size estimates at the national level given that actual data are typically available in only a small number of urban centers where PSE studies, mapping and/or HIV biological and behavioral surveys (IBBS) have been conducted. KPs are often hidden and hard-to-reach and conventional survey and sampling methods do not perform well in these populations; while there are several methods available for size estimation, all are subject to biases and have large levels of statistical error (i.e., large confidence intervals). Beginning in 2009, Indonesia adopted a sophisticated statistical strategy using multivariable regression models to extrapolate estimates based on mapping conducted at locations (“venues”) frequented by KPs in selected districts to other districts where mapping had not been conducted, to arrive at a national PSE. This approach also included additional adjustments to the estimates to account for KP members who may not have been reflected in the mapping figures. Indonesia was one of just 13 countries globally to use regression modeling to produce national estimates.2In 2012, PSEs for KP were estimated using this modeling method based on 2010 mapping data. However, the PSE results may have resulted in under or over estimations because of the questionable quality of mapped data. The final report recommended that future mapping activities follow a more rigorous process using a standardized protocol. In 2014, a standardized mapping exercise was carried out among KPs incorporating these recommendations including assessing mobility or frequency of visits, conducting more than one visit to hot spots, gathering data from non-key population members at the selected venues in addition to KP members, and the inclusion of low- and middle- KP density areas in addition to high density areas.

  

1 UNAIDS/WHO Working Group on Global HIV/AIDS and STI Surveillance. Guidelines on Estimating the Size of

Populations Most at Risk to HIV. 2010. Geneva, Switzerland.

  

2 Ministry of Health of Indonesia. 2012 Size Estimation of Key Affected Populations (KAPs). 2014. Jakarta, Indonesia.

  

3 Sabin K, Zhao J, Garcia Calleja JM, Sheng Y, Arias Garcia S, Reinisch A, et al. Availability and Quality of Size

Estimations of Female Sex Workers, Men Who Have Sex with Men, People Who Inject Drugs and Transgender Women in Low- and Middle-Income Countries. Sandstrom P, editor. PLoS One. 2016. 11(5):e0155150. Available from: http://dx.plos.org/10.1371/journal.pone.0155150.

  

4 Yu D, Calleja JMG, Zhao J, Reddy A, Seguy N. Estimating the Size of Key Populations at Higher Risk of HIV

infection: a Summary of Experiences and Lessons Presented During a Technical Meeting on Size Estimation among Key Populations in Asian Countries. West Pacific Survey Response. 2014;5(3).

  

5 Magnani R, Sabin K, Saidel T, Heckathorn D. Review of sampling hard-to-reach and hidden populations for HIV

surveillance. AIDS 2005;19:S67-S72

  2012 to 2015 mapping in over 70 districts (of 511 districts and 34 provinces nationally), in locations where KPs are known to gather (“venues” or “hotspots”). During the mapping, the number of KPs that frequent each hotspot was estimated by local KP members and stakeholders during meetings. In addition, field teams visited each venue at times of peak attendance to ascertain an observed count of the number of KP members present. While these mapping counts were limited to districts where mapping is conducted, the predictor variables utilized in the extrapolation models were drawn from the national census and health and social surveys, all of which have national coverage, and therefore provide the basis for extrapolating the observed mapping counts to the districts that were not included in the mapping.

  The 2016 PSE process for Indonesia involved many steps, including the assessment of data and the process of obtaining mapping data, selection of predictor variables for the extrapolation models, developing and fitting the regression models to the mapping data, using the fitted models to extrapolate the mapping counts to the unmapped districts, conducting further adjustments to account for KP members who do not attend venues, additional calculations to convert mapping counts to annualized figures that represent the number of KP during a 12-month period, and assessing the resulting size estimates for face validity.

  In March and November of 2016, MOH conducted the latest round of PSE. From May 30 to June 2 2016, technical experts representing WHO, FHI360, UNICEF and UNAIDS met in Bandung, Indonesia with statisticians, methodologists and program managers from the MOH National AIDS Committee (NAC), National Statistics Office, Airlangga and Udayana Universities to provide recommendations to refine the final methodology to ensure the most accurate PSEs possible. Another working group meeting was held in July 2016, with technical experts to review and refine the PSEs. Serial team workshops involving core team, reviewers and contributors were conducted on August, September, and October 2016 to finalize the PSE and accommodate further inputs.

  2 mETHODOLOGy

  2.1. Objectives

  The objectives of estimating the sizes of MSM, FSW, PWID, Waria, and clients of FSW as well clients of Waria are to be able to:

  1. Evaluate the coverage of existing programs

  2. As basis for program planning

  3. Estimate the number of PLHIV in the current year,

  4. Project the number of PLHIV and burden of HIV in the future,

  2.2 General Overview of the PSE process

  There are several steps involved in the process to generate national PSE of KP in Indonesia, all of which are described in detail in this report and in supplemental annexes (Figure 1). This process began with assessing the data from a mapping exercise conducted in 2014 - 2015 (Section 3.1 and Annex). Data from the mapping exercise had been reduced to account for potential double-counting (i.e., a mobility adjustment). The mobility-adjusted estimates provide information about the number of KP members who frequent hotspots during peak times of attendance in the mapped districts. The next step involved using these estimates from the “mapped” districts in the remaining districts that were not mapped by extrapolation using regression modeling (section 3.2). The mapping data (in mapped districts) and extrapolated estimates (in unmapped districts) were adjusted further by an inflation factor in order to reflect those KP members who do not frequent the types of venues where mapping was conducted. These inflation factors were derived from data from the IBBS surveys of FSW, waria and PWID conducted in 2015 (Section 3.3). For MSM, the inflation factor was based on consensus estimates. Finally, local experts reviewed the estimates to assess face validity, including comparing them to other sources of data (i.e., HIV outreach program data, ARV coverage). To some extent, PSE team considered to make further exercise with other inflation factors when most of local experts agreed that the preliminary PSEs were not correct. There was agreement to use PSE result in the provincial level rather in district level. There is also suggestion to provide further explanation on how to use the PSE result wisely (ANNEX).

  D. Adjust PSE D based on review of face validity C C. Apply inflation or deflation factors to account for hidden KP member B A. Begin with mapping counts from hotspots in B. Extrapolate mobility-adjusted mapping counts to unmapped districts using regression model A hotspots (See Mapping Report) mapped districts, adjusted for mobility between

2.3. Definition of KPs

  In a PSE exercise, how the KP is defined has important implications for the interpretation of the estimates and should be determined based on the intended use of the PSEs. To illustrate, Figure 2 presents concentric circles, which represent different subsets of MSM. In this (hypothetical) illustration, the size of each circle is proportional to the relative number of MSM in each subgroup. The figure demonstrates how different definitions of MSM can lead to very different size estimates. For instance, if the objective is to model all potential HIV transmission(circle C) over the course of a year, the MSM “population” could be defined broadly as all men who have had sex with another man in the past 12 months; alternatively, the population could be defined more narrowly, as those MSM who have engaged in high-risk behaviors (i.e., unprotected anal sex) in the past month, if the objective is to plan for short-term prevention programming targeting those currently at greatest risk of acquiring HIV infection (circle E).

  

Figure 2. Defining men who have sex with men according to the objectives of the PSE

  A. Total number of sexually active malesl A

  B. Males who EVER had sex with a male B

  C. Males who have had sex with a male in the PAST YEAR C

  D. Males who have had sex with a males in the PAST MONTH D

  E. Males who have had high-risk sex with a male in the PAST MONTH E

  Generally, different data sources used to calculate the PSE also reflect different subgroups. For instance, the mapping data reflect KPs who frequent known hotspots and are therefore thought to reflect KP members who are currently active in high-risk behaviors; however, the actual time period of the risk behavior, the type of behavior (i.e., oral vs. anal sex), and the age of those mapped are not measured. In contrast, the IBBS surveys reflect KPs who meet survey eligibility criteria, namely, being 15 years or older and having engaged in the respective risk behavior during the last month (for 6 DFSW and IDFSW) or during the last 12 months (for MSM and PWID).

  These differences in definitions are important for interpreting each data source (See Annex 1 for definitions of KP in the mapping and IBBS). The key challenge for developing the PSE is to work with these differences and produce final estimates that reflect the desired segment of the target population. The analysis conducted as a part of this PSE exercise was designed to produce size estimates that reflect the population definitions in Table1.

6. No time frame was used to define eligibility for Waria in the IBBS.

  

Table 1. Definitions of KP for the final size estimates

Population Definition

FSW Females aged 15 years or older who receive money or goods in

  exchange for anal or vaginal penetrative sex in the last 12 months

  

MSM Biological males aged 15 years or older who had sex with a man in

  the last 12 months

  

PWID Males or females aged 15 years or older who injected drugs categorized

  as narcotics in the last 12 months

  

Waria Biological males aged 15 years or older who identify their gender

  identity as females

  

Clients of FSW Males aged 15 to 49 years old who paid a female money or goods for

  anal or vaginal penetrative sex in the last 12 months

  Client of Waria

  Males ages 15 to 49 years old who paid a Waria (as defined above) money or goods for anal penetrative sex in the last 12 months

3 STEPS IN THE CALCULATION OF KPs

3.1. Review of mapping data

  Mapping data collected from 72 to 114 districts per KP (comprising 14% to 22% of all 511 districts nationally) were used to develop the extrapolation models. The MOH protocol for the 2014 and 2015 mapping exercise was also reviewed and the mapping process was discussed with MOH and NAC stakeholders in order to understand the representativeness and reliability of the mapping data. Data from 2014 to 2015 mapping exercise had taken mobility adjustment; but not for the 2012 mapping exercise. For more information on the collection of mapping data, the mobility adjustment and limitations, please see Annex 1.

  

3.2. Extrapolation of mapped counts to unmapped districts using regression

models

  The purpose of the regression models was to extrapolate the mapping findings (i.e., the number of KP members who frequent hotspots in each district), already adjusted for mobility, to districts where mapping had not been conducted. One model was developed per KP. The extrapolation models use data from national demographic and health surveys to predict the mapping findings. Because a wealth of variables is available in these surveys, the first step of extrapolation was to select which variables should be examined for potential inclusion in the regression models (“candidate variables”).

  3.2a. Defining the candidate predictor variables

  Candidate variables for the extrapolation models were selected based on theoretical considerations of what factors could potentially be strongly associated with the number of KPs found at known hotspots. A conceptual framework was developed by technical and local experts at the April - June 2016 meetings in Jakarta and Bandung. The themes (constructs) that were hypothesized to be related to the mapping count of each KP are shown in Table 2. The types of constructs included in the conceptual framework are driven to a large extent by the types of data available in the national surveys; they include for the most part population size, poverty, crime and the types of establishments present in a district or its constituent villages.

  

models

Constructs Theorized association with number of KPs at hotspots MSM Presence of touristic establishments, Such locations are known hotspots (i.e., places such as movie theaters, pubs, where MSM gather and look for partners) discotheques, tennis courts, swimming pools and fitness centers Male employment in services sector MSM are thought to work in particular branches

  Demographics FSW characteristics to be young, divorced, less

  Poverty Waria are thought to have limited economic

  Waria are thought to work at particular branches of the services sector, such as salons, bars, karaoke, hotels and restaurants

  parks, fields, and pavilions Large services sector

  where Waria find clients)

  Waria Presence of touristic establishments, Such locations are known hotspots (i.e., places such as open public areas, hotels,

  the availability of tourist establishment

  

Larger population FSW are thought to often live in larger cities due to

  educated, mobile with respect to work

  to not legal

  of the services sector, such as gymnastic, marketing, and entertainment

  

Legal environment FSW usually related with law repressive action due

  resources and come from poor families

  Poverty FSW are thought to have limited economic

  the services sector, such as salons, bars, karaoke, hotels and restaurants

  parks, fields, and pavilions Female employment in services sector FSW are thought to work particular branches of

  where FSW find clients)

  FSW Presence of touristic establishments, Such locations are known hotspots (i.e., places such as open public areas, hotels,

  the present of touristic establishment

  

Larger population MSM are thought to often live in larger cities due to

  resources, lower income, living in slum areas

  Constructs Theorized association with number of KPs at hotspots

Larger population Waria are thought to often live in larger cities due to

  the availability of touristic establishment

  PWID Crime rates PWID are thought to be more involved in crimes,

  including drug offenses as well as non-drug offenses caused in part by substance abuse

  

Poverty PWID are thought to often have social problems in

  the early life, such as poverty or becoming street children

  

Larger population PWID are thought to often live in larger cities due to

  the availability of the drug supply Data for the candidate predictor variables came from the following data sources:

  1. Sakernas survey (2015): This survey measures labor data from household members, aged ≥10 years old in all 511 districts in Indonesia. A total of 200,000 households 7 were randomly selected.

  2. Susenas survey (2015): This survey measures socioeconomic conditions, including health, education, fertility, family planning, housing, etc. from household members in all 511 districts in Indonesia. The survey is conducted twice per year. The first survey included a sample of 300,000 randomly selected households and was designed to provide district-level estimates. The second survey included a sub-sample of the 8 first survey of 75,000 households to provide province-level estimates.

  3. PODES survey (2014): A survey of 73,709 randomly selected villages in all 511 districts in Indonesia. Provides data related to infrastructure, population and reproductive health, natural resources, and education at the village level. Data were gathered 9 from government officials in the sampled villages, such as the head of the village.

  4. Indonesia Population Projection. Based on national census in year 2010, projection 10 was made to obtain number of population age 15-49 years old in year 2015.

  Most of the predictor variables considered were expressed as proportions by dividing the respective variable (a number of individuals in each district) by the total number of females and/or males aged 15 – 49 years old in the same district, according to population 10 projection data (e.g., proportion of divorced young women among females ages 15 to

  49 years). The denominator included females in models of FSW, males in models of MSM and waria, and both females and males in models of PWID.

  

7. National Statistic Office. Survei Angkatan Kerja Nasional 2015. Available at: https://sirusa.bps.go.id/sirusa/index.php/

dasar/pdf?kd=5&th=2015

  

8. National Statistic Office. Indonesia - Survei Sosial Ekonomi Nasional 2015 Maret (KOR). http://microdata.bps.go.id/

mikrodata /index.php/catalog/657

  

9. National Statistic Office. Indonesia – Pendataan Potensi Desa 2014. http://microdata.bps.go.id/mikrodata/index.php /

catalog/PODES)

10. National Development Planning, National Statistic Office and UNFPA 2013. Proyeksi Penduduk Indonesia 2010 – 2035.

  Available at: http://www.bappenas.go.id/files/5413/9148/4109/Proyeksi_Penduduk_Indonesia_2010-2035.pdf level and thus expressed as a proportion of villages with a given characteristic (e.g., proportion of villages with a bar, proportion of villages with a private high school). These variables could not be disaggregated further.

  The rationale for specifying some variables as proportions rather than absolute numbers is to account for differences in the size (in terms of total population or number of villages) of the districts by modeling the relative level of the given characteristic in each district.

  However, because this step might limit the variation among districts, particularly when counts are very small relative to the total population, we left some potential predictor as absolute numbers (e.g., reported number of males age 15-49 years old). Finally, we explored as potential predictors the total number of males and females ages 15-49 years 10 old in each district, from the population projection , hypothesizing that a district’s size would be related to the number of KP members found at identifiable venues.

  Considering that mapping data were not available in all districts in all region, we also included region as a predictor variable. We defined 6 regions included Sumatera, Jawa, Kalimantan, Bali and NTT, Sulawesi, and Maluku and Papua. The final predictors chosen are displayed in Table 3.

  

Predictors Variabel MSM FSW Waria PWID

  Proportion of villages with income from warehouse sector Prop12 √

  Proportion of villages with slum area Prop16 √ √ √ Proportion of villages with mine industry Prop17 √ Proportion of villages with open public area Prop24 √ √ Proportion of villages per district with functional movie theater Prop25 √ Proportio of villages with functional pub or Prop26 √ discotheque Proportion of villages with function mobile signal phone

  Prop28 √ Proportion of villages with food stall Prop34 √ Proportion of villages with hotel Prop35 √ Proportion of villages with street children location

  Prop41 √ √ Proportion of villages with narcotic cases Prop38 √ Proportion of village with field tennis Prop52 √ Proportion of village with swimming pool Prop53 √ Proportion of villages with fitness center Prop55 √ Proportion of males age 15-49 years old who work at PropXp4 √ service sector Proportion of females age 15-49 years old who move for work in the last year and in the PropXw11 √ service sector Total population age 15-49 years old in the Total_pop √ √ √ district Region Region √ √ √ √

  3.2b. Algorithm to calculate the KPs

  PSE team conducted several steps of statistical analysis to develop the model of mapped district and finally to extrapolate this model to unmapped district. Those steps include: 1) selecting the predictor variable, 2) fitting the best regression model, and 3) applying the regression model for unmapped district to predict the final estimates of KP. These steps are presented in Figure 3. After variables were selected based on theoretical model, the simple regression model were applied to select variables which have significant correlation (P value <0.05) and higher adjusted R-square. Three models of Linear, Poisson and Negative Binomial were chosen to run the regression as the type of dependent variable is a count. This step resulted three functional forms of regression. Each predictor variable for KP which was significant in bivariate regression was included in the multiple regression model. In each of the three-functional form model, the relationship was tested using different transformations of the predictor: natural log, square root and polynomial (x + x2).

  B.

  Fitting the best fit model Following the multiple regression using the Linear, Poisson and Negative Binomial regression, test of classical assumption was done to determine the best regression model; by assessing the goodness of fit and the predictive result of KP number. Based on those criteria, the best regression model for modeling all KP was Linear Regression Multiple Form.

  We measured the goodness of fit from Multivariable Regression Model using AIC correction and Adjusted R squared. Classical assumption for the Linear regression model were also tested, including test for:

  • autocorrelation (by assessing Durbin Watson value),
  • homoscedasticity (by assessing the plots of standardized regression and heteroscedasticity val
  • linearity (by assessing the plots of predicted values), multicollinearity (by assessing the VIF and tolerance) to obtain the best-fit model, and - normality (by assessing skewness and kurtosis plots (Z-score)).

  C.

  Applying the regression model for unmapped district to predict the final estimates of KP Remaining variables that were found to no longer be significant were removed from the multivariable model through a stepwise backward regression with a threshold of P<0.10. Ninety-five percent confidence intervals, and in some cases plausibility bounds (i.e., when the lower confidence bound was lower than the mapped number), were calculated based on the standard errors as part of procedure to predicting the KPs number using STATA’s command.

  Once the predictions were made, a variable was created representing the initial PSE estimates for 1) mapped districts, using the actual mobility-corrected mapping count and 2) unmapped districts, using the predicted mobility-corrected mapping count. The final regression model is presented in Table 4.

  Data setting of KPs based on PODES, SUSESNAS and SAKERNAS Getting bivariate regression for each KP in 3 functional regression form

  Negative Binomial Bivariate Regression Form Poisson Regression Bivariate Form

  Modelling Multiple Regression using each KP predictor (from bivariate regression) Linear Regression Bivariate Form

  Negative Binomial Multiple Regression Form Poisson Regression Multiple Form

  Linear Regression Multiple Form Grouping and setting variable of each KP from the data source

  Evaluating each selected variable by assessing significant correlation (pvalue), and adj. R 2 Selecting predictor variable using simple Linear, Poisson and Negative

  Binomial regression Y i = ( y i + i ) y i ! ( i ) i ( μ i + i ) i μ i ( μ i + i ) y i ln( μ i ) = i = + 1 X i + 2 X i 2 + ... +

k

X

ik

1 2 Yi 2 ... i i i X X β β β ε = + + + +

  1. Assessing Goodness of Fit (Correlation number and Significantly predictors of KPs)

  2. Assessment of Predictive Result (R-squared)

  3. Adjustment regarding to KPs estimation number Selecting Multivariate Regression Model à Linear Regression Multiple Form (Measuring Goodness of Fit from Multivariable Regression Model based on AIC correction and Adjusted R squared)

  Classical Assumption Linear Regression Testing (Autocorrelation test, Homoscedasticity test, Linearity, Multicolinearity, and Normality) Predicting the number of KPs (final estimation for each KP)

  KP Regression Equation R AIC Point 95% CI Square Lower Upper

FSW 0,71 1713,5 138.101 32.320 222.476

  Y = 1079,9 prop16 + 537,4 prop24

  (n= 114) + 8,3 prop34 + 42,1 prop35 –

  2382,9 prop41 – 1148644xw11 – 415,3 dummy2 +0,82 total_pop_sqrt

  

MSM 0,21 1585,8 124.626 310.427

  Y = 2033,5 prop_laki + 76,1

  (n=103) prop25 + 16,8 prop 26 – 461,8

  prop52 – 154,5 prop53 + 31,4 prop 55 – 4233,9 xp4 + 155,3 dummy2 – 397,7 dummy6

  Waria

  Y = 31,5 prop12 + 47,5 prop16 – 0,24 1192,4 33.378 76.599

  (n=95) 71,8 prop17 – 11,2 prop24 +

  99,2 prop28 + 0,1 total_pop

  PWID

  Y = 258,5 prop16 – 185,5 prop38

   (n=72)

  • 398,8 prop41 + total_pop 0,21 937,4 26.557 67.394

3.3. Using inflation factors to account for non-venue-going KP

  Based on review of the types of venues included in the mapping data, we applied adjustments to account for KPs who do not frequent these types of venues (“hidden population”), and therefore were not likely to have been captured during mapping. These were developed from responses from the IBBS surveys conducted in 2015 (Table 5). For more information on the collection of IBBS data and limitations of the IBBS data, please see Annex 1.

  Based on those questions, we calculated proportion of survey participants who have answer “YES” to the option that reflected potentially missing in mapping for each KP. The number of these participants were divided by total IBBS participants then resulting proportion of missing population in mapping. These proportions were used as an inflation factors for each KP in the district level based on the general rules (Table 6). For instance, when the percent of hidden population of FSW was 19.5%, then the number obtained from the regression model will be considered as only 81.5% of the true size of FSW population in the district. Therefore, there will be added another 19.5% x total KP obtained from the regression for respective district. It is also applied for all KPs in the district. Instead the inflation factor, we also applied the deflation factor (Q610) for FSW to account for migration between cities, because during the mapping it was not corrected.

   PWID MSM FSW Waria Q858. Q309. Q610. Q310.

  Have you met/know Where do you most Prior to this city, Where do you an outreach worker often hang out with in how many city most often hang of HIV & STI? other MSM have you sell sex? out with peers? Q867.

  In the past year, have you attend meetings/discussion with health worker regarding with HIV?

  Rationale: Rationale: Rationale: Rationale:

  to account if someone to account for Q610 is used to to account for has been someone account for mobility someone who reached/exposed to who do not go to across do not go to HIV prevention meeting venues city/districts meeting venues

  (initially Q615 was used, but the calculation resulting PSE <200.000) Q867 is used to account if someone has been reached/exposed to HIV prevention.

  • * Also used same question in the 2013 IBBS data for district without mapping in 2015 (FSW = Q608/Q803; PWID

    = Q854; MSM = Q952)

  

Availabilty of IBBS data for the district How adjustment for “hidden population”

members was determined Available for district District’s own IBBS estimate No IBBS in district, but available from Average estimate from district in the same other districts in same province province No IBBS in entire province, but available Average estimate from districts in the same from other districts in same region region No IBBS in entire region* Average estimate from all districts in the IBBS