Itimate parter violece ad HIV i te sub-sahara Africa coutries: what do the Demographic ad Health Surveys tell us? Dick Durevall, Aika Lidskog Summary Backgroud May studies have idetified a sigificat positive relatio betwee itimate parter violece ad HIV i wome, but adjusted aalyses have produced icosistet results. We systematically assessed the associatio, ad uder what coditio it holds, usig atioally represetative data from te sub-sahara Africa coutries, focusig o physical, sexual, ad emotioal violece, ad o the role of male cotrollig. Methods We assessed cross-sectioal data from 12 Demographic ad Health Surveys from te coutries i sub- Sahara Africa. The data are atioally represetative for wome aged 15 49 years. We estimated odds ratios usig logistic regressio with ad without cotrols for demographic ad socioecoomic factors ad survey regio fixed effects. Exposure was measured usig physical, sexual, emotioal violece, ad male cotrollig, ad combiatios of these. The samples used were ever-married wome, married wome, ad wome i their first uio. Depedig o specificatio, the sample size varied betwee 11 231 ad 45 550 wome. Fidigs There were cosistet ad strog associatios betwee HIV ifectio i wome ad physical violece, emotioal violece, ad male cotrollig (adjusted odds ratios raged from 1 2 to 1 7; p values raged from <0 0001 to 0 0058). The evidece for a associatio betwee sexual violece ad HIV was weaker ad oly sigificat i the sample with wome i their first uio. The associatios were depedet o the presece of cotrollig ad a high regioal HIV prevalece rate; whe wome were exposed to oly physical, sexual, or emotioal violece, ad o cotrollig, or whe HIV prevalece rates are lower tha 5%, the adjusted odds ratios were, i geeral, close to 1 ad isigificat. Iterpretatio The fidigs idicate that male cotrollig i its ow right, or as a idicator of ogoig or severe violece, puts wome at risk of HIV ifectio. HIV prevetio itervetios should focus o high-prevalece areas ad me with cotrollig, i additio to violece. Lacet Glob Health 2015; 3: e34 43 Published Olie November 21, 2014 http://dx.doi.org/10.1016/ S2214-109X(14)70343-2 See Commet page e4 Departmet of Ecoomics ad Gotheburg Cetre of Globalizatio ad Developmet, Uiversity of Gotheburg, Gotheburg, Swede (Prof D Durevall PhD, A Lidskog PhD) Correspodece to: Dr Aika Lidskog, Departmet of Ecoomics, School of Busiess, Ecoomics ad Law, Uiversity of Gotheburg, PO Box 640, SE 405 30, Gotheburg, Swede aika.lidskog@ecoomics. gu.se Fudig Swedish Natioal Sciece Foudatio ad Gotheburg Cetre of Globalizatio ad Developmet, Uiversity of Gotheburg, Gotheburg, Swede. Copyright Durevall et al. Ope Access article distributed uder the terms of CC BY-NC-SA. Itroductio Accordig to the UNAIDS World AIDS Day Report 2011, 1 oe i seve ew HIV ifectios i South Africa could have bee avoided through the prevetio of itimate parter violece. The statemet is based o Jewkes ad colleagues logitudial study 2 of youg wome i South Africa i 2002 06. Several other studies also show a associatio betwee itimate parter violece ad HIV ifectio i wome, ad a recet report from WHO ad UNAIDS 3 cocludes that research documets a udeiable lik betwee itimate parter violece ad HIV ifectio. However, a multicoutry study 4 with atioally represetative data ot oly questios whether itimate parter violece causes HIV but eve suggests that o associatio exists. The study used six Demographic ad Health Surveys (DHS) from sub-sahara Africa ad three from other developig coutries (Domiica Republic, Haiti, ad Idia). Additioally, a recet literature survey 5 cocluded that, although several studies report a positive ad statistically sigificat associatio betwee exposure to itimate parter violece ad HIV ifectio, the fidigs are icosistet ad the type of itimate parter violece that is related to HIV remais uclear. The aim of this report is to aalyse systematically the associatio betwee HIV ad itimate parter violece with use of all sub-sahara Africa DHS datasets available i early 2014, ad to assess uder what coditios the associatio is recorded. Our exposures of itimate parter violece are biary idicators of physical, sexual, ad emotioal violece (deoted violece for simplicity), male cotrollig, ad combiatios of these. Most studies focus o some of these factors, mostly physical ad sexual violece, ad oly oe study that used DHS data aalysed cotrollig, 6 although such is viewed as oe type of itimate parter violece. 7 www.thelacet.com/lacetgh Vol 3 Jauary 2015 e34
See Olie for appedix Cotrollig is likely to be importat both i its ow right ad as a proxy for severity of violece. However, it ofte occurs i combiatio with physical or sexual violece that is ogoig ad ulikely to stop. 8 Violece i combiatio with cotrollig has bee termed coercive cotrollig violece, to distiguish it from situatioal couple violece that results from coflicts betwee parters that occasioally escalate ito violece. 9 Coercive cotrollig violece is likely to result i frequet abuse ad severe ijuries, whereas situatioal couple violece is more commo but is probably ot related to male domiace sice ofte both me ad wome are aggressors. If the distictio betwee coercive cotrollig violece ad situatioal couple violece matters, measures of itimate parter violece that igore cotrollig might be oly weakly associated with HIV. We use a arrow measure of cotrollig : positive resposes to questios about whether the husbad does ot allow his wife to meet her girlfrieds or tries to limit her cotact with family. Our hypothesis was that a associatio exists betwee itimate parter violece ad HIV, ad that this associatio is stroger whe violece is combied with cotrollig. By cotrast with earlier studies that used DHS data, we esure that compariso groups are clea ie, wome i the compariso groups do ot report beig exposed to ay of the four types of itimate parter violece. If a woma does ot belog to the clea compariso group, ad is ot exposed to the type of violece uder cosideratio i the estimatio, we exclude her from the sample. Earlier studies aalysed whether wome exposed to oe type of itimate parter violece are more likely to be HIV positive tha are a compariso group of wome that icludes those who might have bee exposed to aother type of itimate parter violece (ie, the compariso group is ot clea). The samples of wome used i earlier studies based o DHS data vary betwee ever-married wome, 4 married wome, 4,6,10 ad wome who are i their first uio ad do ot report ay premarital sex or extramarital sex withi the previous 12 moths. 11 Argumets both for ad agaist ay of the three choices exist: may ever-married wome could have bee ifected by HIV after divorce or durig widowhood, or they might have divorced a violet husbad who ifected them; ad faithful wome i their first uio are least likely to have bee ifected outside marriage, but premarital or extramarital sex might be misreported. We aimed to compare the three sample groups systematically. HIV prevalece varies greatly across coutries ad regios i sub-sahara Africa, from lower tha 1% (eg, i rural Mali) to more tha 20% (eg, i urba Zimbabwe) i adults i our sample. As is evidet from the potetial mechaisms geeratig the associatio betwee itimate parter violece ad HIV, the associatio might be stroger i societies with a high prevalece of HIV. Three fudametal, potetially importat, mechaisms for this associatio exist: 12 violet me are more likely tha o-violet me to become ifected by HIV outside their marriage; wome exposed to itimate parter violece are more likely to be ifected outside their marriage tha are those ot exposed to such violece (either because her risky triggers itimate parter violece, or because itimate parter violece leads to reduced self-esteem ad icreased risky sexual ); ad itimate parter violece icreases the risk of HIV trasmissio betwee parters because of uwated sex ad less codom use, icludig coercio ad subsequet geital trauma. If the first two mechaisms are importat, prevalece should matter, whereas it should ot be of major sigificace for the associatio if icreased trasmissio risk withi marriage is the clearly domiat mechaism. Therefore, we explore how the associatios vary across regios with high ad low levels of HIV prevalece. Methods Data collectio We used cross-sectioal DHS data from 12 surveys i te coutries: oe each from Burkia Faso (2010), Côte d Ivoire (2011/12), Gabo (2011), Keya (2008), Liberia (2006), Mali (2007), Rwada (2005), ad Zambia (2007), ad two from Malawi (2004 ad 2010), ad Zimbabwe (2005/06 ad 2010/11). These datasets comprise all Africa DHS data available i early 2014 that icluded HIV testig ad a complete domestic violece module ad that allow us to lik HIV ad itimate parter violece (ie, by askig the same wome questios about both). We did ot use four surveys with data for HIV ad itimate parter violece because of missig iformatio: emotioal violece is missig i Rwada 2010; cotrollig is missig i Keya 2003 ad Rwada 2010; ad, i Cameroo 2011, iformatio about HIV ad itimate parter violece was gathered from differet subgroups of wome. The DHS data are atioally represetative for wome 15 49 years of age. The surveys are stratified by coutry-specific admiistrative ad geographical regios. Detailed iformatio about survey desig, samplig methods, ad refusal rates is available i the DHS fial survey reports. 13 Of the eligible sample, typically more tha 90% participated i the survey. The respose rate for the itimate parter violece module was higher tha 98% i most surveys; the exceptio is Rwada 2005, i which it was 89%. The respose rate for HIV testig varied betwee 76% ad 99%. Appedix p 1 provides iformatio about missig observatios ad a aalysis of likely biases. Outcome ad exposure The HIV outcome is a biary variable idicatig HIV ifectio status (ie, positive or egative). Blood spot samples were collected from every idividual i the same radom subset of households where me were eligible for iterviews, o a fully iformed ad volutary basis. The e35 www.thelacet.com/lacetgh Vol 3 Jauary 2015
blood samples were first aalysed with a ELISA test, the all positive samples ad 10% of the egative oes were retested with ELISA. For discordat samples, a Wester blot test was fially used. 14 Exposure to itimate parter violece is measured by biary (ie, yes or o) idicators of physical, sexual, ad emotioal violece, cotrollig, ad combiatios of these. Iformatio about itimate parter violece was collected from oe radomly selected woma i each household, with o-oe else i the household aware that this was doe. Married or cohabitig wome were asked about ever havig experieced itimate parter violece by their husbad or parter, whereas formerly married or formerly cohabitig wome were asked about itimate parter violece by their most recet husbad or parter. The violece module used is a modified ad abbreviated versio of the Coflict Tactics Scale, 15 i which the questios ask about specific acts (eg, does/did your husbad ever slap you, puch you with his fist, twist your arm, etc), which are classified as physical, emotioal, or sexual violece (see appedix pp 4 5 for details). If the woma aswered yes to at least oe of the questios related to physical, sexual, or emotioal violece, the relevat violece idicator was coded as 1; if she aswered o to all questios i the set, it was coded 0. Wome were also asked specific questios about marital cotrol exercised by their curret husbad (or their most recet husbad if they were widowed or divorced), such as whether he is jealous or agry if she talks to other me, or whether or ot he allows her to meet her female frieds. Cotrollig was defied as at least oe positive respose to questios about whether the husbad does ot permit her to meet her girlfrieds or tries to limit her cotact with her family. Although jealousy ad accusatios of ifidelity are likely to be commo traits i cotrollig me, this iformatio was ot used because it could also be related to the woma s ifidelity, which itself could be a trigger of itimate parter violece. Jealousy ad suspicios of ifidelity are reported two to three times more ofte by the wome i our sample tha are attempts to limit cotact with female frieds ad family. For the sample of ever-married wome, the Spearma s rak correlatio coefficiets betwee jealousy or ifidelity ad restrictios o seeig female frieds or family rage from 0 20 to 0 35, depedig o the sample group, so the idicator we used is measurig somethig other tha jealousy or suspicio of ifidelity. I additio to the idividual idicators of physical, sexual, ad emotioal violece ad cotrollig, we combied idicators to assess the importace of male cotrollig : ay violece (physical, sexual, or emotioal) or cotrollig ; ay violece but o cotrollig ; cotrollig but o violece; ay violece plus cotrollig ; physical violece but o cotrollig ; sexual violece but o cotrollig ; ad emotioal violece but o cotrollig. We also used a idicator that combies physical or sexual violece, which was also used i Harlig ad colleagues study 4 the most comprehesive previous study of atioal data from Africa coutries. To help comparisos with their results, we estimated models with this idicator with both a clea compariso group ad a compariso group that icludes wome exposed to emotioal violece or cotrol, i additio to those who have ot bee exposed to ay itimate parter violece. Details about the codig of each idicator ca be foud i appedix pp 4 5. Covariates The covariates used i this report are stadard ad similar to the oes used i the multicoutry study by Harlig ad colleagues, 4 with the exceptio of lifetime umber of sex parters, which was excluded from our study because it could be a mediatig factor. The covariates are age, educatio, occupatio, religio, wealth quitile, ad urba residece. Estimatios with all ever-married wome also icluded a marital status idicator. I pooled sample estimatios, we used survey-specific dummies for withicoutry regios because HIV rates vary geographically ad over time; Harlig ad colleagues 4 used coutry dummies. Statistical aalysis We estimated bivariate ad multivariate logistic models usig pooled samples ad samples from each survey, varyig the exposure idicator ad keepig the referece group clea ie, the estimatio sample oly icludes wome exposed to itimate parter violece as measured i that model ad a cotrol group of wome who were ot exposed to ay form of itimate parter violece. We estimated models usig ever-married wome, married wome, ad wome who are curretly i their first uio ad do ot report ay premarital or extramarital sex durig the previous 12 moths. Importatly, these samples are ot mutually exclusive. Samples also vary depedig o the itimate parter violece idicator used. Although the clea compariso group is always the same, the group of wome subjected to itimate parter violece differs depedig o the type of violece. Furthermore, to show the importace of regioal HIV prevalece i explaiig the lik betwee itimate parter violece ad HIV, samples were split ito subatioal regios with HIV prevalece higher tha 5% ad those with rates below 5%, which is roughly the average prevalece i sub-sahara Africa. These results are preseted i appedix pp 11 12. Stadard errors are clustered at the regio survey level by use of the Eicker-Huber-White robust variace estimator, which assumes idepedece across groups but allows ay type of withi-group correlatio. We followed Harlig ad colleagues approach 4 ad did ot use weights i the mai estimatios because their effects are uclear whe subsamples are used. We used Stata versio 13.0 for all aalyses. www.thelacet.com/lacetgh Vol 3 Jauary 2015 e36
Role of the fudig source The fuders of the study had o role i study desig, data collectio, data iterpretatio, data aalysis, writig of the report, or the decisio to submit for publicatio. The correspodig author had full access to all the data i the study ad had fial resposibility for the decisio to submit for publicatio. Results Table 1 shows the pooled sample size ad percetage distributio by outcome, exposure, ad covariates for all ever-married wome. I total, 39 000 observatios with o-missig data for all variables are listed. Almost 40% of these wome report havig bee exposed to some form of itimate parter violece (termed ay violece ). This percetage varies i the coutry surveys from 26% i Burkia Faso to 63% i Zambia (appedix p 6). Physical violece is the most commo type (29%), Wome, (%) HIV-positive wome, (%) Age group (years) 15 19 2469 (6 3%) 118 (4 8%) 20 24 7806 (20 0%) 595 (7 6%) 25 29 8765 (22 5%) 1029 (11 7%) 30 34 7303 (18 7%) 1079 (14 8%) 35 39 5397 (13 8%) 792 (14 7%) 40 44 3961 (10 2%) 505 (12 7%) 45 49 3299 (8 5%) 341 (10 3%) Marital status Curretly married 34 497 (88 5%) 3098 (9 0%) Formerly married 4503 (11 5%) 1361 (30 2%) Urbaicity Rural 28 254 (72 4%) 2884 (10 2%) Urba 10 746 (27 6%) 1575 (14 7%) Wealth quitiles Poorest 8860 (22 7%) 794 (9 0%) Poorer 8238 (21 1%) 829 (10 1%) Middle 7812 (20 0%) 859 (11 0%) Richer 7918 (20 3%) 1092 (13 8%) Richest 6172 (15 8%) 885 (14 3%) Educatio No educatio 12 731 (32 6%) 544 (4 3%) Primary 16 040 (41 1%) 2084 (13 0%) Secodary or higher 10 229 (26 2%) 1831 (17 9%) Employmet Not employed 13 062 (33 5%) 1779 (13 6%) Agricultural 13 351 (34 2%) 979 (7 3%) Maual 2238 (5 7%) 325 (14 5%) Other 10 349 (26 5%) 1376 (13 3%) Religio Christia 28 067 (72 0%) 3870 (13 8%) Muslim 8383 (21 5%) 314 (3 7%) Other religio 2550 (6 5%) 275 (10 8%) (Table 1 cotiues o ext colum) followed by emotioal violece (23%), whereas sexual violece is less frequet (12%). Although cotrollig is defied arrowly, as may as 22% of wome report havig bee exposed to it. The prevalece of HIV ifectio is 14 16% i the wome who report itimate parter violece ad 10% i those who report o violece (table 1). The distributio of HIV prevalece across the other covariates is i lie with fidigs i previous studies: it icreases with age util 40 years (ad the declies) ad also icreases with risig wealth ad educatio; it is much higher amog formerly married wome (ie, those who are widowed or divorced) tha i curretly married wome; ad it is higher i urba tha rural areas (table 1). Table 2 reports uadjusted ad adjusted odds ratios based o pooled regressios for 13 measures of itimate parter violece ad the three samples: ever-married wome; married wome; ad wome i their first marriage who do ot report premarital or extramarital sex. All the uadjusted odds ratios are higher tha 1 ad early all are sigificat. The adjusted odds ratios are also higher tha 1 ad may, but ot all, are sigificat. Physical violece, emotioal violece, ad cotrollig as idividual factors are associated with a Wome, (%) HIV-positive wome, (%) (Cotiued from previous colum) Violece Ay violece 15 204 (39 0%) 2078 (13 7%) Physical violece 11 188 (28 7%) 1542 (13 8%) Sexual violece 4608 (11 8%) 714 (15 5%) Emotioal violece 9043 (23 2%) 1317 (14 6%) Cotrollig 8730 (22 4%) 1155 (13 2%) No violece or cotrollig 20 376 (52 2%) 2023 (9 9%) Coutry Burkia Faso, 2010 4834 (12 4%) 65 (1 3%) Côte d Ivoire, 2011/12 2268 (5 8%) 108 (4 8%) Gabo, 2011 2681 (6 9%) 193 (7 2%) Keya, 2008 2195 (5 6%) 233 (10 6%) Liberia 2006 2617 (6 7%) 59 (2 3%) Mali, 2007 2746 (7 0%) 44 (1 6%) Malawi, 2004 2082 (5 3%) 327 (15 7%) Malawi, 2010 5020 (3 9%) 687 (13 7%) Rwada, 2005 2476 (6 3%) 82 (3 3%) Zambia, 2007 3384 (8 7%) 609 (18 0%) Zimbabwe, 2005/06 4154 (10 7%) 1034 (24 9%) Zimbabwe, 2010/11 4543 (11 6%) 1018 (22 4%) Total 39 000 (100 0%) 4459 (11 4%) The sample i this table is ever-married wome for whom complete iformatio was available. Table 1: Pooled sample size ad percetage distributio by exposure ad covariates e37 www.thelacet.com/lacetgh Vol 3 Jauary 2015
icreased risk of HIV ifectio i all three sample groups, with adjusted odds ratios ragig from 1 12 to 1 42 ad p values ragig from less tha 0 0001 to 0 0058, whereas sexual violece is isigificat (accordig to adjusted odds ratios) i the samples of ever-married ad married wome but sigificat i the sample of wome i their first marriage (p=0 0361). The combied idicators that iclude cotrollig all have statistically sigificat adjusted odds ratios, whereas the associatio is much weaker whe o cotrollig is ivolved. For example, the adjusted odds ratios for cotrol ad o violece for the three samples are 1 21 (95% CI 1 05 1 39) for ever-married wome, 1 25 (1 08 1 45) for married wome, ad 1 31 Ever-married wome Married wome Wome i their first uio with o premarital or extramarital sex Uadjusted Uadjusted Uadjusted Uadjusted Uadjusted Uadjusted Physical violece 32 495 1 424* (1 236 1 641) 31 072 1 133 (1 037 1 237) 28 784 1 375* (1 168 1 619) 27 475 1 222* (1 096 1 363) 16 638 1 686* (1 432 1 986) 15 283 1 423* (1 232 1 643) Sexual violece Emotioal violece Cotrollig Cotrollig ad emotioal violece Violece or cotrollig Violece ad cotrollig Cotrollig ad o violece Violece ad o cotrollig Physical violece ad o cotrollig Sexual violece ad o cotrollig Emotioal violece ad o cotrollig Physical or sexual violece (a) Physical or sexual violece (b) 25 583 1 653* (1 383 1 976) 30 318 1 512* (1 286 1 777) 29 977 1 354* (1 172 1 564) 24 773 1 687* (1 399 2 035) 40 179 1 344* (1 188 1 521) 26 431 1 564* (1 317 1 856) 24 406 1 040 (0 867 1 248) 31 062 1 336* (1 161 1 538) 28 087 1 339* (1 170 1 534) 23 538 1 499* (1 233 1 823) 26 405 1 391* (1 161 1 667) 33 956 1 423* (1 238 1 635) 45 550 1 263* (1 114 1 433) 24 432 1 077 (0 960 1 209) 29 073 1 122 (1 034 1 218) 28 718 1 201* (1 102 1 308) 23 664 1 215* (1 084 1 361) 38 589 1 104 (1 026 1 188) 25 253 1 192 (1 071 1 326) 23 282 1 208 (1 054 1 386) 29 919 1 032 (0 939 1 134) 26 898 1 093 (0 988 1 209) 22 460 1 023 (0 888 1 179) 25 236 1 047 (0 944 1 162) 32 510 1 104 (1 010 1 208) 43 820 1 054 (0 969 1 146) 22 900 1 512* (1 232 1 857) 26 866 1 498* (1 249 1 796) 26 686 1 328* (1 122 1 573) 22 099 1 675* (1 351 2 077) 35 559 1 311* (1 133 1 516) 23 505 1 530* (1 263 1 855) 22 295 1 059 (0 840 1 335) 27 987 1 296 (1 096 1 532) 25 360 1 291 (1 100 1 516) 21 379 1 373 (1 087 1 734) 23 881 1 389 (1 135 1 700) 30 047 1 362* (1 157 1 603) 40 247 1 201 (1 041 1 386) 21 691 1 136 (0 976 1 322) 25 716 1 243* (1 120 1 380) 25 196 1 351* (1 222 1 494) 20 782 1 507* (1 308 1 735) 34 101 1 163 (1 062 1 272) 22 117 1 412* (1 238 1 609) 20 952 1 247 (1 075 1 446) 26 924 1 042 (0 929 1 169) 24 252 1 112 (0 986 1 253) 20 234 0 984 (0 812 1 193) 22 793 1 097 (0 969 1 242) 28 716 1 166 (1 042 1 305) 38 652 1 089 (0 979 1 211) 13 531 1 739* (1 342 2 253) 15 447 1 659* (1 348 2 042) 15 337 1 534* (1 246 1 889) 12 903 1 862* (1 451 2 390) 20 275 1 518* (1 298 1 775) 13 612 1 858* (1 463 2 360) 13 229 1 154 (0 881 1 512) 16 442 1 505* (1 262 1 794) 14 986 1 550* (1 331 1 804) 12 772 1 500 (1 119 2 012) 14 048 1 550* (1 227 1 958) 17 324 1 668* (1 413 1 969) 23 642 1 458* (1 251 1 699) 12 136 1 239 (1 014 1 514) 13 854 1 222 (1 070 1 394) 14 111 1 407* (1 215 1 629) 11 322 1 380* (1 147 1 662) 19 168 1 253* (1 127 1 393) 12 163 1 493* (1 221 1 827) 11 726 1 306 (1 075 1 588) 14 954 1 150 (1 015 1 302) 13 572 1 318* (1 129 1 539) 11 231 1 036 (0 812 1 322) 12 399 1 129 (0 976 1 306) 15 946 1 355* (1 180 1 557) 22 414 1 309* (1 139 1 503) The values i this table are based o logistic regressios. Regressios for the adjusted s are cotrolled for age, educatio, occupatio, religio, wealth, urba residece, ad survey regio dummies. Marital status is cotrolled for i the ever-married samples. The referece group is always wome exposed to o violece of ay type or to cotrollig. Wome exposed to violece or cotrollig that was ot captured by the exposure idicator were excluded from the estimatio sample, except for physical or sexual (b), i which the sample icludes wome who report beig exposed to emotioal violece ad cotrollig. Stadard errors were clustered at the survey regio level. Note that estimatio samples ca be larger or smaller tha the 39 000 wome (table 1) because of differeces i data availability. For estimatio of adjusted s, the sample sizes are reduced because observatios from subatioal regios with o HIV-ifected wome are excluded from the regressios. =odds ratio. *p<0 001. p<0 05. p<0 01. Clea compariso group. Compariso group icludes wome exposed to emotioal violece ad cotrol. Table 2: Risk of HIV ifectio i wome exposed to itimate parter violece www.thelacet.com/lacetgh Vol 3 Jauary 2015 e38
Physical violece Sexual violece Emotioal violece Cotrollig Burkia Faso, 2010 Côte d Ivoire, 2011 12 Gabo, 2011 Keya, 2008 Liberia, 2006 Uadjusted Uadjusted 4060 1 384 (0 649 2 952) 1771 1 167 (0 560 2 433) 1837 0 804 (0 605 1 067) 1582 2 004* (1 020 3 937) 2463 1 022 (0 481 2 170) Mali, 2007 2097 1 302 (0 830 2 042) Malawi, 2004 Malawi, 2010 Rwada, 2005 Zambia, 2007 Zimbabwe, 2005 06 Zimbabwe, 2010 11 1351 1 161 (0 613 2 198) 3416 1 036 (0 879 1 221) 1994 1 048 (0 606 1 814) 2441 1 193 (0 859 1 657) 2683 1 189 (0 920 1 537) 3089 1 271* (1 020 1 585) 4013 1 172 (0 529 2 597) 1761 1 100 (0 579 2 088) 1826 0 823 (0 587 1 153) 1449 1 776 (0 692 4 558) 1795 1 043 (0 533 2 044) 1721 1 232 (0 653 2 323) 1346 1 297 (0 722 2 328) 3416 1 032 (0 894 1 192) 1975 1 274 (0 667 2 433) 2434 1 072 (0 830 1 385) 2673 1 299 (0 969 1 742) 3066 1 375 (1 182 1 600) Uadjusted Uadjusted Uadjusted Uadjusted NA NA NA NA 4068 1 536 (0 900 2 621) 1381 1 433 (0 564 3 642) 1061 1 014 (0 679 1 513) 1177 2 115* (1 032 4 335) 1627 0 827 (0 374 1 828) 1757 2 120* (1 059 4 244) 1222 1 074 (0 475 2 429) 3234 0 781 (0 509 1 200) 1546 1 538 (0 693 3 410) 1583 1 313 (0 785 2 194) 2135 0 934 (0 684 1 276) 2580 1 093 (0 788 1 515) 1374 1 316 (0 618 2 803) 1052 1 015 (0 644 1 601) 1097 1 319 (0 475 3 663) 1208 1 325 (0 587 2 988) 1411 2 135 (0 857 5 315) 1216 1 448 (0 680 3 087) 3234 0 884 (0 579 1 348) 1264 1 289 (0 550 3 023) 1577 1 164 (0 780 1 737) 2128 1 076 (0 760 1 523) 2562 1 185 (0 869 1 616) 1649 0 958 (0 446 2 056) 1522 0 785 (0 552 1 117) 1453 1 783* (1 013 3 137) 2470 1 189 (0 555 2 545] 1920 0 486 (0 181 1 306) 1169 1 186 (0 456 3 082) 3606 0 942 (0 783 1 133) 1492 2 073 (0 964 4 460) 1802 1 449* (1 087 1 932) 2739 1 360 (1 158 1 598) 2976 1 298 (1 078 1 563) 4015 1 355 (0 826 2 222) 1641 0 848 (0 427 1 686) 1510 0 786 (0 549 1 125) 1450 1 387 (0 647 2 973) 1800 1 082 (0 616 1 900) 1555 0 453 (0 123 1 666) 1164 1 307 (0 604 2 825) 3606 0 940 (0 754 1 171) 1476 2 149 (0 950 4 859) 1796 1 216 (0 966 1 531) 2730 1 467 (1 248 1 725) 2954 1 349 (1 153 1 579) Uadjusted Uadjusted 4139 1 201 (0 545 2 645) 1668 1 576 (0 921 2 698) 1692 0 919 (0 604 1 398) 1344 1 541 (0 786 3 022) 2283 1 274 (0 478 3 392) 2386 0 772 (0 443 1 345) 1524 1 340 (0 879 2 043) 3117 1 030 (0 741 1 433) 1708 1 481 (0 795 2 760) 1925 1 489 (1 145 1 937) 2264 1 386 (1 207 1 591) 2636 1 456 (1 205 1 760) 4090 1 109 (0 563 2 185) 1660 1 395 (0 808 2 410) 1679 0 892 (0 595 1 338) 1151 1 223 (0 481 3 110) 1664 1 170 (0 522 2 621) 1980 0 604* (0 379 0 963) 1519 1 477 (1 098 1 985) 3117 1 080 (0 871 1 339) 1550 1 211 (0 623 2 353) 1920 1 383 (1 133 1 687) 2254 1 536 (1 323 1 784) 2612 1 614 (1 294 2 012) The values i this table are based o logistic regressios. Regressios for the adjusted s are cotrolled for age, educatio, occupatio, religio, wealth, ad urba residece. Marital status is cotrolled for i the ever-married samples. The referece group is always wome exposed to o violece of ay type ad to o male cotrollig. Wome exposed to violece or cotrol that was ot captured by the exposure idicator were excluded from the estimatio sample. Stadard errors are clustered at the survey regio level. For estimatio of adjusted s, the sample sizes are reduced because observatios from subatioal regios with o HIV-ifected wome are excluded from the regressios. =odds ratio. NA=ot applicable (o variatio i sample). *p<0 05. p<0 01. p<0 001. Table 3: Risk of HIV ifectio i married wome exposed to itimate parter violece, by survey ad type of violece (1 08 1 59) for wome i their first marriage, whereas for ay violece but o cotrollig these odd ratios are lower at 1 03 (95% CI 0 94 1 13), 1 04 (0 93 1 17), ad 1 15 (1 02 1 30), respectively (table 2). Oly oe adjusted odds ratio is statistically sigificat for aother idicator of itimate parter violece i which the wome do ot report experiecig male cotrollig : physical violece i the sample of wome i their first uio (table 2). Whe physical ad sexual violece are combied, as i Harlig ad colleagues 2010 study, 4 ad the cotrol groups iclude wome affected by emotioal violece ad cotrollig (ie, they are ot clea), the adjusted odds ratios are isigificat ad close to 1: 1 05 (95% CI 0 97 1 15) for ever-married wome, ad 1 09 (0 98 1 21) for married wome. Whe a clea cotrol group is used istead, the adjusted odds ratios are sigificat: 1 10 (95% CI 1 01 1 21) for ever-married wome ad 1 17 (1 04 1 31) for married wome (table 2). I the sample of wome i their first uio, the adjusted odds ratio is greater tha 1 ad is sigificat i both models. The associatio betwee itimate parter violece ad HIV is oticeably weaker whe the samples are limited to idividual surveys, ad may odds ratios are isigificat (tables 3, 4). The adjusted odds ratios i the samples with married wome are sigificat ad higher tha 1 for physical violece i Zimbabwe 2010 11, for emotioal e39 www.thelacet.com/lacetgh Vol 3 Jauary 2015
Violece or cotrollig Violece ad o cotrollig Cotrollig ad violece Cotrollig ad o violece Uadjusted Uadjusted Uadjusted Uadjusted Uadjusted Uadjusted Uadjusted Uadjusted Burkia Faso, 2010 Côte d Ivoire, 2011 12 Keya, 2008 Liberia, 2006 4723 1 068 (0 644 1 770) 2104 1 291 (0 735 2 269) 1910 1 681 (0 972 2 905) 3193 1 112 (0 569 2 171) Mali, 2007 2735 0 925 (0 630 1 359) Malawi, 2004 Malawi, 2010 Rwada, 2005 Zambia, 2007 Zimbabwe, 2005 06 Zimbabwe, 2010 11 1877 1 194 (0 792 1 800) 4277 0 897* (0 814 0 990) 1 101 (0 643 1 885) 2889 1 221 (0 913 1 633) 3426 1 178 (0 999 1 388) 3806 1 209* (1 008 1 449) 4661 0 918 (0 577 1 460) 2093 1 132 (0 667 1 922) 1906 1 432 (0 701 2 925) 2343 0 981 (0 547 1 757) 2302 0 803 (0 571 1 130) 1871 1 355* (1 010 1 818) 4277 0 913 (0 809 1 030) 2269 1 123 (0 645 1 955) 2881 1 130 (0 916 1 394) 3415 1 281 (1 074 1 528) 3773 1 271 (1 086 1 488) 4118 0 930 (0 327 2 644) 1722 1 046 (0 513 2 135) 1482 1 788* (1 050 3 045) 2249 0 945 (0 580 1 539) 2017 1 244 (0 679 2 279) 1321 0 975 (0 589 1 614) 3742 0 837 (0 652 1 076) 1841 0 813 (0 468 1 410) 2095 1 014 (0 681 1 510) 2828 1 076 (0 839 1 381) 3190 1 086 (0 842 1 401) 4067 0 696 (0 328 1 474) 1712 0 963 (0 520 1 783) 1478 1 405 (0 667 2 959) 1678 0 966 (0 386 2 416) 1646 1 225 (0 551 2 723) 1315 1 154 (0 756 1 761) 3742 0 843 (0 640 1 109) 1823 0 981 (0 521 1 844) 2087 0 937 (0 681 1 291) 2821 1 147 (0 864 1 524) 3166 1 121 (0 893 1 408) 3730 1 867 (0 608 5 733) 1474 1 297 (0 557 3 017) 1203 1 760 (0 752 4 119) 1962 1 473 (0 520 4 173) 1868 0 926 (0 500 1 715) 1174 1 286 (0 407 4 066) 2953 1 041 (0 711 1 525) 1502 1 665 (0 839 3 308) 1687 1 436* (1 085 1 901) 2101 1 547 (1 280 1 871) 2455 1 484 (1 299 1 695) 3688 1 648 (0 599 4 534) 1467 1 107 (0 522 2 348) 948 1 324 (0 434 4 036) 1432 1 524 (0 718 3 236) 1509 0 747 (0 410 1 363) 1169 1 516 (0 530 4 340) 2953 1 071 (0 799 1 436) 1356 1 427 (0 617 3 300) 1682 1 264* (1 004 1 590) 2091 1 720 (1 427 2 075) 2437 1 637 (1 388 1 930) 3943 0 885 (0 349 2 246) 1480 1 852 (0 682 5 032) 1057 1 110 (0 665 1 855) 1660 0 892 (0 309 2 571) 2186 0 712 (0 260 1 954) 1318 1 373 (0 940 2 004) 2746 1 006 (0 719 1 407) 1460 1 262 (0 596 2 675) 1369 1 615 (1 175 2 219) 1829 0 995 (0 605 1 634) 2201 1 392 (0 860 2 253) 3898 0 843 (0 395 1 798) 1472 1 823 (0 620 5 360) 891 0 830 (0 376 1 831) 1237 0 659 (0 214 2 031) 1795 0 543 (0 210 1 402) 1313 1 452* (1 062 1 985) 2746 1 080 (0 997 1 170) 1187 1 051 (0 489 2 259) 1364 1 658 (1 273 2 158) 1823 1 080 (0 634 1 842) 2180 1 599 (0 939 2 723) The values i this table are based o logistic regressios. Regressios for the adjusted s are cotrolled for age, educatio, occupatio, religio, wealth, ad urba residece. Marital status is cotrolled for i the ever-married samples. The referece group is always wome exposed to o violece of ay type ad to o male cotrollig. Wome exposed to violece or cotrol that was ot captured by the exposure idicator were excluded from the estimatio sample. Stadard errors were clustered at the survey regio level. For estimatio of adjusted s, the sample sizes are reduced because observatios from subatioal regios with o HIV-ifected wome are excluded from the regressios. =odds ratio.*p<0 05. p<0 01. p<0 001. Table 4: Risk of HIV ifectio i married wome exposed to itimate parter violece, by survey ad combied measures of itimate parter violece violece i the two Zimbabwea surveys, ad for cotrollig i Malawi 2004, Zambia 2007, ad the two Zimbabwea surveys (table 3); the exceptio is cotrollig i Mali where the odds ratio is sigificat ad lower tha 1. The results are similar whe idicators are combied, ad, as whe the data are pooled, violece without cotrollig has little effect (table 4). The results are also similar for ever-married wome ad wome i their first uio who do ot report premarital or extramarital sex, although adjusted odds ratios for physical violece are sigificat i several surveys (appedix pp 7 10). Because most of the sigificat odds ratios are i coutries with a high prevalece of HIV (eg, Zimbabwe has the highest HIV prevalece of the icluded coutries), we split the pooled sample ito subatioal regios with HIV rates higher tha ad lower tha 5% to assess the role of HIV prevalece. I the sample with HIV prevalece higher tha 5%, there are 56 regios, whereas i the sample with HIV prevalece lower tha 5% there are 73 regios. Re-estimatio of the models with HIV rates above 5% produces essetially the same results as previously, whereas, i regios with HIV rates lower tha 5%, oly four adjusted odds ratios are sigificatly higher tha 1 (appedix pp 11 12). Discussio Our aalysis of pooled DHS data from 12 surveys ad te sub-sahara Africa coutries cofirms that reported itimate parter violece is associated with a sigificatly raised risk of HIV ifectio i wome (pael). Depedig o the exposure idicator, the sample (ever-married wome, married wome, ad wome i their first marriage who report o premarital or extramarital sex), www.thelacet.com/lacetgh Vol 3 Jauary 2015 e40
Pael: Research i cotext ad regioal HIV prevalece, the adjusted odds ratios rage from 1 1 to 1 7, ofte with p values less tha 0 0001. The choice of exposure idicator has a substatial effect o the results. Male cotrollig ad physical ad emotioal violece geerally icrease the probability of HIV ifectio, whereas sexual violece is sigificat oly i the sample of wome i their first uio. Cotrollig has a key role: whe ay violece (physical, sexual, or emotioal) is ot combied with cotrollig, adjusted odds ratios geerally do ot differ sigificatly from 1. The differeces betwee the adjusted odds ratios for violece that is ot Systematic review We did ot do a systematic review of the scietific literature because several comprehesive surveys exist. 5,16,17 The most recet survey 5 foud 101 studies about the associatio betwee itimate parter violece ad HIV, of which 45 use data from Africa. Most of these studies are cross-sectioal studies that use small samples, ad oly a small proportio are populatio-based. Six studies used atioal survey data from various sub-sahara Africa coutries. 4,6,10,11,18,19 Noe reported sigificat adjusted odds ratios for physical ad sexual violece; two showed that emotioal violece was associated with HIV, usig the Rwada 2005 Demographic ad Health Survey; 6,11 oe showed a associatio betwee a combied measure of physical, sexual, ad emotioal violece ad HIV, usig the Keya 2008 09 Demographic ad Health Survey; 10 ad oe showed o effect of sexual power o HIV i South Africa. 19 Four studies about sub-sahara Africa used logitudial data. 2,20 22 Two are from Rakai i Ugada, oe of which reported that wome exposed to itimate parter violece are 55% more likely tha other wome to get ifected withi 1 year, 20 whereas the other 22 showed o statistically sigificat associatio. A third study 21 did ot show that itimate parter violece icreases the risk of ifectio i a sample of HIV-discordat couples i easter ad souther Africa, although already ifected wome were more likely to report itimate parter violece tha those who were ot ifected. The fourth study is Jewkes ad colleagues report 2 of youg wome i South Africa, which reported that oe i seve HIV ifectios were attributable to itimate parter violece. Iterpretatio Our results corroborate evidece from much of the published literature of a associatio betwee itimate parter violece ad HIV ifectio, ad the size of the odds ratios (1 1 1 7) are i lie with those of Jewkes ad colleagues study. 2 However, they differ from earlier studies that used the atioally represetative Demographic ad Health Survey data. Three reasos exist for these differeces: we use idicators of physical, sexual, ad emotioal violece combied with male cotrollig ; we compare wome exposed to a particular type of itimate parter violece versus wome who report ot beig exposed to ay type of itimate parter violece; ad we estimate models with data from regios with high ad low levels of HIV prevalece. The associatio is strogest i high-prevalece regios whe the woma is exposed to male cotrollig, whereas there is almost ever a associatio for violece that is ot combied with cotrollig. Oe iterpretatio of this fidig is that our idicator of cotrollig measures male domiace that is associated with frequet ad ogoig violece, whereas violece without cotrollig is more ofte the result of quarrels i which both spouses might be aggressors. However, more research is eeded to uderstad the role of cotrollig : it could be a idicator of severe violece or importat i its ow right. I the meatime, HIV prevetio programmes should focus o itimate parter violece i high-prevalece areas, with a particular focus o male cotrollig. combied with cotrollig versus violece that is combied with cotrollig or cotrollig that is ot combied with violece are statistically sigificat. Moreover, adjusted odds ratios for idividual measures of physical, sexual, ad emotioal violece are ot sigificat whe wome exposed to cotrollig are excluded, with oe exceptio: physical violece i the sample of wome i their first uio. Oe iterpretatio of these results is that cotrollig combied with other types of itimate parter violece is a proxy for cotrollig coercive violece. Measures of violece without cotrollig should therefore to a large extet measure acts classified as situatioal couple violece. However, cotrollig might also coceivably capture severe violece, although the statistically sigificat odds ratios of cotrollig but o violece suggest that it measures other effects ad it might be importat i its ow right. Idepedet of iterpretatio, our defiitio of cotrollig seems to have empirical relevace. Whe idividual surveys are aalysed, may of the adjusted odds ratios are o-sigificat, although most of them are greater tha 1 (table 3, appedix pp 7 10). Use of the sample weights provided by DHS does ot affect the results much (appedix pp 13 14). The statistically sigificat associatios betwee itimate parter violece ad HIV ted to be recorded i regios with a high HIV prevalece; i particular, the associatio is strog for Zimbabwe, where more tha 20% of married wome are HIV positive. This fidig suggests that the failure to fid a associatio i may surveys could be related to the prevalece level, which is supported by the aalysis of the sample split ito subatioal regios with HIV ifectio rates higher tha ad lower tha 5% (appedix pp 11 12). Malawi, a coutry i which prevalece varies greatly across its three regios, provides a good example: testig the regios separately shows several sigificat associatios i the souther regio (where the HIV ifectio rate is 17 6%), but few i the cetral regio (6 5%) ad orther regio (8 1%) (appedix p 15). The importace of HIV prevalece poits to mechaisms that lik itimate parter violece ad HIV through icreased ifectio risk of oe or both spouses outside of marriage, by cotrast with icreased risk of trasmissio betwee spouses caused by itimate parter violece. 12 Moreover, may observatios might be eeded whe prevalece is low, sice low variatio i the depedet variable makes estimates less precise: as metioed, very few poit estimates of the adjusted odds ratios are lower tha 1. After all, the HIV epidemic is dyamic: may factors affect the risk of beig ifected by the virus. Additioally, collectio of data about itimate parter violece ad HIV ifectio is challegig, ad we have little iformatio about the cotext i which itimate parter violece occurs or how regular it is i a give relatioship. e41 www.thelacet.com/lacetgh Vol 3 Jauary 2015
Several reasos exist for the icosistet fidigs i the earlier studies, three of which are addressed i this study. First, the correlatio betwee the idicators of itimate parter violece is ofte igored, so wome exposed to oe type of itimate parter violece are compared with a group i which some wome are exposed to other types of itimate parter violece; thus, the cotrol group is ot clea. Secod, the idicators of violece used cotai both ogoig severe violece that is a expressio of male domiace ad violece resultig from coflicts betwee parters, i which both parters might be aggressors. Although our measures are arrow ad far from perfect, we seem to be able to reduce some of this measuremet problem by icludig male cotrollig. Third, as previously discussed, the probability of fidig a associatio betwee HIV ad itimate parter violece depeds o the level of HIV prevalece i the commuity. Our results differ from those of Harlig ad colleagues 4 because they used cotrol groups that icluded wome who report emotioal violece ad cotrollig, ad focused o samples with either ever-married wome or married wome. Use of this approach with the six Africa surveys icluded i their study, or with all our surveys, results i isigificat adjusted odds ratios. However, whe the cotrol group is clea (as i our study), a statistically sigificat associatio exists betwee physical or sexual itimate parter violece ad HIV ifectio. As is the case with all observatioal studies, especially those ito sesitive topics, this study has limitatios. The use of cross-sectioal data makes it challegig to cotrol for relevat cofouders. Noetheless, may of the potetial cofouders that we did cotrol for are highly sigificat, ad substatial differeces exist betwee uadjusted ad adjusted odds ratios. Furthermore, the icluded cofouders are all ulikely to be affected by itimate parter violece (appedix p 16). Aother limitatio of cross-sectioal data is that we do ot kow whether HIV ifectio occurred after the acts of itimate parter violece or vice versa. Not everyoe eligible for HIV testig could or wated to participate i the surveys, so a potetial selectio bias exists. The aalysis of missig data i appedix pp 1 3 suggests that estimated associatios betwee itimate parter violece ad HIV could be biased dowwards, but this coclusio is oly tetative. Measuremet error could also affect the estimated associatio dowwards by creatig atteuatio bias. Because we did tests usig several idicators ad may differet samples, multiple testig, with a icreased probability of obtaiig false-positive results (type I errors), could be a cocer. However, our samples ad idicators are ot idepedet, so adjustmet of sigificace levels is ot straightforward. Use of the Boferoi correctio as a approximatio gives a sigificace level of roughly 0 001, but it assumes o true positives, that the tests are idepedet, ad that oly oe false-positive result is accepted at the 0 05 level for all tests. 23 Therefore, the correctio makes it far too difficult to reject the ull hypothesis, but provides a idicatio that we have several true positives. Moreover, the study is set up to compare associatios uder differet coditios, ad we do ot base our coclusios o sigle test outcomes, but o the geeral patter. This study provides additioal support for the view that itimate parter violece is associated with icreased risk of HIV ifectio by showig a strog associatio betwee the two variables i DHS data from various Africa coutries. It also emphasises the importace of male cotrollig for the associatio physical, sexual, ad emotioal violece without cotrollig seem to be of a differet kid of itimate parter violece ad shows that the associatio probably depeds o HIV prevalece i the eighbourhood. It therefore presets evidece i favour of HIV prevetio programmes that focus o reducig itimate parter violece i high-prevalece areas. Cotributors DD ad AL both aalysed the data ad wrote the report. Declaratio of iterests We declare o competig iterests. Ackowledgmets The study is based o aoymous public data with o idetifiable iformatio about the survey participats. Access to the HIV testig datasets required a olie ackowledgemet of the coditios of use by MEASURE DHS. We thak the referees, Guilla Kratz, the participats of the STRIVE semiar at the Lodo School of Hygiee ad Tropical Medicie (Lodo, UK), ad the Nordic Coferece of Developmet Ecoomics 2013 for useful commets. We also thak the Swedish Research Coucil ad the Gotheburg Cetre of Globalizatio ad Developmet, Uiversity of Gotheburg (Gotheburg, Swede), for fudig. Refereces 1 UNAIDS. World AIDS Day Report 2011. Geeva: Joit Uited Natios Programme o HIV/AIDS, 2011. 2 Jewkes RK, Dukle K, Ndua M, Shai N. Itimate parter violece, relatioship power iequity, ad icidece of HIV ifectio i youg wome i South Africa: a cohort study. Lacet 2010; 376: 41 48. 3 WHO. Addressig violece agaist wome ad HIV/AIDS: what works? Geeva: World Health Orgaizatio ad UNAIDS, 2010. 4 Harlig G, Msisha W, Subramaia SV. No associatio betwee HIV ad itimate parter violece amog wome i 10 developig coutries. PLoS Oe 2010; 5: 1 8. 5 Kouyoumdjia FG, Fidlay N, Schwadt M, Calzavara LM. A systematic review of the relatioships betwee itimate parter violece ad HIV/AIDS. PLoS Oe 2013; 8: 1 25. 6 Kayibada JF, Bitera R, Alary M. Violece toward wome, me s sexual risk factors, ad hiv ifectio amog wome: fidigs from a atioal household survey i Rwada. J Acquir Immue Defic Sydr 2012; 59: 300 07. 7 WHO. Prevetig itimate parter ad sexual violece agaist wome. Geeva: World Health Orgaizatio, 2010. 8 Johso MP. Coflict ad cotrol: geder symmetry ad asymmetry i domestic violece. Violece agaist wome. 2006; 12: 1003 18. 9 Kelly JB, Johso MP. Differetiatio amog types of itimate parter violece: research update ad implicatios for itervetios. Fam Court Rev 2008; 46: 476 99. 10 Shi C-F, Kouyoumdjia F, Dushoff J. Itimate parter violece is associated with HIV ifectio i wome i Keya: a cross-sectioal aalysis. BMC Public Health 2013; 13: 1 7. www.thelacet.com/lacetgh Vol 3 Jauary 2015 e42
11 Dude A. Spousal itimate parter violece is associated with HIV ad other STIs amog married Rwada wome. AIDS Behav 2011; 15: 142 52. 12 Dukle KL, Decker MR. Geder-based violece ad HIV: reviewig the evidece for liks ad causal pathways i the geeral populatio ad high-risk groups. Am J Reprod Immuol 2013; 69: 20 26. 13 The DHS Program. DHS Coutry Fial Reports. http:// dhsprogram.com/publicatios/publicatio-search.cfm?type=5 (accessed Oct 1, 2014). 14 Macro Iteratioal. HIV prevalece estimates from the Demographic ad Health Surveys. Calverto: Macro Iteratioal, 2008. 15 Kisho S. Domestic violece measuremet i the demographic ad health surveys: the history ad the challeges. Expert paper for UN Divisio for the Advacemet of Wome, Geeva, 2005. http://www.u.org/womewatch/daw/egm/vaw-stat-2005/docs/ expert-papers/kishor.pdf (accessed Nov 10, 2014). 16 Adersso N, Cockcroft A, Shea B. Geder-based violece ad HIV: relevace for HIV prevetio i hyperedemic coutries of souther Africa. AIDS 2008; 22: S73 86. 17 Shamu S, Abrahams N, Temmerma M, Musekiwa A, Zarowsky C. A systematic review of Africa studies o itimate parter violece agaist pregat wome: prevalece ad risk factors. PLoS Oe 2011; 6: e17591. 18 Mattso CL, Settergre S, Sabatier J. Spousal sexual violece, HIV, ad sexually trasmitted ifectios: a evaluatio of Demographic ad Health Survey data Zimbabwe (2005 2006), Malawi (2004), ad Keya (2003). Am J Epidemiol 2009; 169: S1 137. 19 Pettifor AE, Measham DM, Rees HV, Padia NS. Sexual power ad HIV risk, South Africa. Emerg Ifect Dis 2004; 10: 1996 2004. 20 Kouyoumdjia FG, Calzavara LM, Body SJ, et al. Itimate parter violece is associated with icidet HIV ifectio i wome i Ugada. AIDS 2013; 27: 1331 38. 21 Were E, Curra K, Delay-Moretlwe S, et al. A prospective study of frequecy ad correlates of itimate parter violece amog Africa heterosexual HIV serodiscordat couples. AIDS 2011; 25: 2009 18. 22 Zablotska I, Gray R, Koeig M, et al. Alcohol use, itimate parter violece, sexual coercio ad HIV amog wome aged 15 24 i Rakai, Ugada. AIDS Behav 2009; 13: 225 33. 23 Abdi H. The Boferoi ad Šidák correctios for multiple comparisos. I: Salkid NJ, ed. Ecyclopedia of measuremet ad statistics. Thousad Oaks: Sage Research Methods, 2007. e43 www.thelacet.com/lacetgh Vol 3 Jauary 2015