. 2020 Jun 18;23(6):e25490. doi: 10.1002/jia2.25490

Table 1.

Randomised controlled trials and observational studies of circumcision and HIV incidence in men by study type, setting and approximate date order of implementation

Study (first author publication year)	Setting and location	Period	HIV infections [person‐years] (incidence per 100 py)†	Incidence ratio (95% CI)	Notes
A) Randomised controlled trial
Auvert 2005 [4]	Orange Farm, Gauteng, South Africa	Jul‐2002 – Apr‐2005	Circ. 20 [2354] (0.85) Not circ. 49 [2339] (2.11)	mITT ^a 0.40 (0.24 to 0.68) ‘As treated’ ^b 0.24 (0.14 to 0.44)	1546 men randomized to immediate and 1582 to delayed circumcision
Bailey 2007 [5]	Nyanza Province, Kenya	Feb‐2002 – Dec‐2006	Circ. 22 [2084 ^c ] (1.06 ^d ) Not circ. 47 [2214 ^c ] (2.12 ^d )	mITT 0.41 (0.24 to 0.70) ‘As treated’ 0.40 (0.23 to 0.68)	1391 men randomized to immediate and 1393 to delayed circumcision
Gray 2007 [6]	Rakai District, Uganda	Aug‐2002 – Dec‐2006	Circ. 22 [3352] (0.66) Not circ. 45 [3392] (1.33)	mITT 0.43 (0.24 to 0.75) ‘As treated’ 0.40 (0.23 to 0.70)	2474 men randomized to immediate and 2522 to delayed circumcision
Pooled			Circ. 64 [7791] (0.82) Not circ.141 [7945] (1.77)	mITT 0.41 (0.30 to 0.56) ^e	Absolute difference: ^f 10 fewer (from 8 to 12 fewer) per 1000 py Heterogeneity statistics: I ² = 0.0%, χ² ₍₂₎ = 0.03, p = 0.98
B) Extended follow‐up of former RCT participants
Mehta 2013 [18]	Nyanza Province, Kenya	Feb‐2002 – Sep‐2010	Circ. 47 [5744 ^c ] (0.82) ^g Not circ. 79 [4107 ^c ] (1.92) ^g	Crude 0.38 (0.26‐0.55) Adj. 0.42 (0.26 to 0.66) ^h	Estimated crude IR from 2 to 6 years 0.36 (0.22 to 0.61) computed as weighted difference between crude IR over full 6‐year period and “as treated” IR over 2 years from RCT (0.40 [0.23 to 0.68])[5] ⁱ
Gray 2012 [19]	Rakai District, Uganda	Dec‐2006 – Dec‐2010	Circ. 48 [9628] (0.50) Not circ. 21 [1087] (1.93)	Crude 0.26 (0.15 to 0.43) Adj. 0.27 (0.16 to 0.45) ^j
Pooled			Circ. 95 [15,373] (0.62) Not circ.100 [5194] (1.93)	Adj. 0.34 (0.24 to 0.49) ^k	Absolute difference ^f : 13 fewer (from 10 to 15 fewer) per 1000 py Heterogeneity statistics: I ² = 35.4%, χ² ₍₁₎ = 1.55, p = 0.21
C) Cohorts of men at high risk of HIV infection
Cameron 1989 [20]	STI clinic patients, Nairobi, Kenya	Mar‐1986 – Dec‐1987	Circ. 6 [60.8 ^l ] (9.9) Not circ. 18 [20.7 ^l ] (87.1)	Crude0.11 (0.04 to 0.30) ^m Adj. 0.12 (0.04 to 0.33) ⁿ	Circumcision prevalence 214/293 (73%)
Lavreys 1999 [21]	Trucking company employees, Mombasa, Kenya	Mar‐1993 – Jun‐1997	Circ. 32 [1280 ^o ] (2.5) Not circ. 11 [186 ^o ] (5.9)	Crude 0.43 (0.23 to 0.91) Adj. 0.25 (0.12 to 0.53) ^p	Circumcision prevalence 651/746 (87%)
Gray 2000 [22]	Serodiscordant couples, Rakai District, Uganda	Nov‐1994 – Oct‐1998	Circ. 0 [106] (0.0) Not circ. 40 [239](16.7)	Crude 0.00 (0.00 to 0.22) ^q Adj. ‐	Circumcision prevalence 50/187 (27%)
Reynolds 2004 [23]	STI clinic patients, Pune, Maharashtra, India	May‐1993 – 2000	Circ. 2 [285] (0.7) Not circ.165 [3013] (5.5)	Crude 0.13 (0.02 to 0.47) Adj. 0.15 (0.04 to 0.62) ^r	Circumcision prevalence 191/2298 (8%)
Hughes 2012 [24]	Serodiscordant couples, 7 countries southern and eastern Africa ^s	Nov‐2004 – Oct‐2008	Circ. 20 [1712] (1.2) Not circ. 24 [1297] (1.9)	Crude 0.53 (0.29 to 0.96) Adj. 0.53 (0.29 to 0.96) ^t	Circumcision prevalence 1225/2223 (55%). Number of infections and py from [25]. ART use in female partner reported in 543 of 12,966 quarterly visits (4%) [25].
Pooled			Circ. 60 [3444] (1.7) Not circ.258 [4756] (5.4)	Adj. 0.29 (0.19‐0.43) ^k	Absolute difference ^f : 39 fewer (from 31 to 44 fewer) per 1000 py Heterogeneity statistics: I ² = 67.0%, χ² ₍₄₎ = 12.1, p = 0.017
D) Community‐based cohorts before circumcision scale‐up
Gray 2000 [22]	Rakai District, Uganda	Nov‐1994 – Oct‐1998	Circ. 18 [1683] (1.07) Not circ.154 [8548] (1.80)	Crude0.59 (0.34 to 0.97) ^q Adj. 0.53 (0.33 to 0.87) ^u	Proportion circumcised 908/5516 (16%).
Shaffer 2007 [26]	Kericho, Rift Valley Province, Kenya	Jun‐2003 – Dec‐2006	Circ. 17 [2165] (0.79) Not circ. 13 [524] (2.48)	Crude 0.31 (0.15 to 0.64) Adj. 0.34 (0.16 to 0.73) ^v	Proportion circumcised 1108/1378 (80%)
Kim 2016[27]	Kenya (national)	Aug‐2006 – Dec‐2007	Circ. 23 [5618 ^w ] (0.41) Not circ. 4 [858 ^w ] (0.47)	Crude 0.88 (0.30 to 2.53) ^x Adj. Not reported	Cross‐sectional survey – recent HIV infection (approximately last 12 months) using Limiting Antigen Avidity Enzyme Immunoassay (LAg) assay. Proportion circumcised 5618/6476 (87%). National ART prevalence 40% in 2007 [28].
Dandona 2013 [29]	Guntur District, Andhra Pradesh State, India	2004 – Jun‐2011	Circ. 1/744 ^y (‐) Not circ. 22/3265 ^y (‐)	Crude 0.20 (0.03 to 1.47) ^z Adj. 0.07 (0.01 to 0.83) ^aa	Proportion circumcised 744/4009 (19%)
Pooled			Circ. 58 [9466] (0.61) ^bb Not circ. 171 [9930] (1.72) ^bb	Adj. 0.48 (0.33 to 0.70) ^k	Absolute difference ^f : 9 fewer (from 5 to 12 fewer) per 1000 py Heterogeneity statistics: I ² = 33.9%, χ² ₍₃₎ = 4.54, p = 0.21
E) Community‐based cohorts during circumcision scale‐up
Grabowski 2017 [30]	Rakai District, Uganda	Apr‐1999 – Aug‐2016	Circ. 97 [16,256] (0.60) Not circ.283 [25,893] (1.09)	Crude 0.54 (0.43 to 0.69) Adj. 0.62 (0.48 to 0.79) ^cc	Circumcision prevalence increased from 15% in 1999 to 59% in 2016 and ART coverage from 12% to 68% among HIV positive adults from 2006 to 2016 (from 8% to 61% in men and from 13% to 72% in women)
Lissouba 2011 [31]	Orange Farm, Gauteng, South Africa	Oct‐2007 – Apr‐2008	Circ. 6 [273 ^o ] (2.2) Not circ. 37 [661 ^o ] (5.6)	Crude 0.40 (0.16 to 0.98) Adj. 0.35 (0.14 to 0.89) ^dd	Baseline survey. Recent HIV infection (last 15 months) using BED assay in men ages 22 to 34 yr using BED assay. Proportion circumcised (clinical examination) 199/698 (29%).Estimated 16% national ART coverage in women in 2007 [32].
Auvert 2013 [33]	Orange Farm, Gauteng, South Africa	Oct‐2010 – Jun‐2011	Circ. ‐ [‐] (1.2) Not circ.‐ [‐] (3.9)	Crude 0.43 (0.25 to 0.69) ^ee Adj. 0.41 (0.23 to 0.70) ^ff	Follow‐up survey. Recent HIV infection (last 12 months) using BED assay (using cut‐off 1.51, correction‐1, corresponding to an approximate 12‐month window for new infections; alternative parameters with different assay windows gave comparable results [see Table 5 in Auvert 2013]). Proportion circumcised 1771/3338 (53%). Estimated 46% national ART coverage in women in 2009 [32].
Vandormael 2019 [34]	Hlabisa, KwaZulu‐Natal, South Africa	2009 – 2017	Circ. 89 ^gg [7840 ^gg ] (1.14) Not circ.429 ^gg [16,932 ^gg ] (2.54)	Crude 0.45 (0.35 to 0.57) ^q Adj. 0.58 (0.47 to 0.71) ^hh	From 2009 to 2016 circumcision prevalence increased from 3% to 33% and ART coverage from 18% to 37% in men and 19% to 49% in women.
Borgdorff 2018 [35]	Siaya County, Nyanza Province, Kenya	Oct‐2010 – Sep‐2016	Circ. 18 [3936 ⁱⁱ ] (0.46) Not circ. 73 [10,459 ⁱⁱ ] (0.70)	Crude 0.66 (0.37 to 1.11) ^q Adj. Not reported	Proportion circumcised 1211/4429 (27%). ART prevalence in women 126/296 (43%). Person‐years of follow‐up estimated assuming average follow‐up time was the same for circumcised and uncircumcised men and corresponded to the mean follow‐up for all men in the cohort. No information on circumcisions performed during study period. Indirect evidence from Kenya Demographic and Health Survey 2014 reported circumcision prevalence in Nyanza Province increased from 45% in 2008 to 72% in 2014 [36].
Kagaayi 2019 [37]	Fishing Communities, Lake Victoria, Uganda	Nov‐2011– Feb‐2017	Circ. 53 [3635] (1.46) Not circ. 69 [1926] (3.58)	Crude 0.41 (0.28 to 0.59) ^q Adj. 0.46 (0.32 to 0.67) ^jj	Circumcision prevalence increased from 35% to 65% and ART coverage from 16% to 82% among HIV‐positive adults from 2011 to 2016 (13% to 78% in men and 18% to 85% in women).
Pooled			Circ. 216 [28,233] (0.77) ^kk Not circ.953 [61,553] (1.55) ^kk	Adj. 0.56 (0.49 to 0.64) ^k	Absolute difference ^f : 7 fewer (from 6 to 8 fewer) per 1000 py Heterogeneity statistics: I ² = 0.0%, χ² ₍₅₎= 4.35, p = 0.50

^{^†}

‘‐’, Not reported; ‘Circ.’, Circumcised; ‘Not circ.’, Not circumcised.

^{^a}

Modified intention‐to‐treat analysis, excluding men subsequently identified as HIV‐positive at enrolment

^{^b}

Analysis according to actual circumcision status

^{^c}

Computed from estimated annualized incidence and number of infections

^{^d}

Annualized incidence estimated from reported cumulative 2‐year life table incidence [P ₁ = 1 ‐ (1 – P ₂)^(1/2)]

^{^e}

Weighted average of log IRs with weights inversely proportional to variance back‐transformed to ratio scale

^{^f}

Computed from pooled IR, confidence interval and incidence in uncircumcised men

^{^g}

Annualized incidence estimated from reported cumulative 6‐year life table incidence [P ₁ = 1 ‐ (1 ‐ P ₆)^(1/6)]

^{^h}

Adjusted for baseline age and time‐varying sexual behaviours and sexually transmitted infections

^ⁱ

Computed as weighted difference of log IRs with weights inversely proportional to variance back‐transformed to ratio scale

^{^j}

Adjusted for sociodemographic characteristics and time‐varying sexual behaviours

^{^k}

Weighted average of log adjusted IRs with weights inversely proportional to variance back‐transformed to ratio scale

^{^l}

Computed from mean follow‐up (weeks) by seroconversion and circumcision status

^{^m}

Computed from number of infections and py at risk

^ⁿ

Odds ratio, adjusted for frequency of contact with prostitutes (once, two or more) and genital ulcer disease

^{^o}

Computed from reported incidence and number of infections

^{^p}

Adjusted for occupation (driver or assistant vs. mechanic or ancillary worker), religion, frequency of extramarital sex

^{^q}

Computed from number of infections and py at risk (exact confidence interval)

^{^r}

Adjusted for fixed (religion, education, living with family) and time‐varying (calendar year, age, marital status, multiple sex partners, sex‐worker partners, condom use, tattoos, medical injections) risk factors

^{^s}

Botswana, Kenya, Rwanda, South Africa, Tanzania, Uganda, Zambia

^{^t}

Adjusted for fixed (age and HSV‐2 status at enrolment) and time‐varying (genital ulcer disease, and partner plasma viral load and condom use) risk factors

^{^u}

Adjusted for randomisation arm, age, marital status, extramarital sex partners, sexually transmitted infection diagnoses

^{^v}

Adjusted for age, education and religion

^{^w}

Number of recent infections from published paper Tables 1 and 2 reversing labels ‘Circumcised’ and ‘Not circumcised’ which had been switched in error [personal communication Andrea Kim, December 2018]. Person‐years computed assuming 12‐month window for recent infection

^{^x}

Computed from number of infections and py at risk (approximate confidence interval from log IR)

^{^y}

Number of HIV infections/number at risk (incidence not reported)

^{^z}

Odds ratio computed from number of infections and number at risk

^{^aa}

Odds of HIV infection adjusted for standard of living, occupation, spouse HIV status, lifetime sex partners, and condom use in past 6 months

^{^bb}

Number of HIV infections and py from African studies only

^{^cc}

Adjusted for survey round, 5‐year age groups, education, sex partners in past year, sex with partners outside community, self‐reported genital ulcer disease in past year, self‐reported non‐marital partnership and consistent condom use, type of community (agrarian, trading), and community HIV prevalence

^{^dd}

Adjusted for age, ethnic group, marital status, lifetime sexual partners, sexual partners in past year, consistent condom use with non‐spousal partners and HSV‐2 status

^{^ee}

Weighted by circumcision propensity score from logistic regression on age, ethnic group, religion, ever fathered child, occupation, age at first intercourse, alcohol consumption, education level, and ever married

^{^ff}

Additionally adjusted for age, lifetime sexual partners, and number of and consistent condom use with non‐spousal partners in the last 12 months

^{^gg}

Personal communication Alain Vandormael January 2020

^{^hh}

Adjusted for age, marital status, household income, condom use, out‐migration, female HIV prevalence and female ART coverage

^ⁱⁱ

Estimated from number at risk and mean follow‐up per man (3.25 py)

^{^jj}

Adjusted for survey round, age, marital status, education and sexual behaviour (sex partners, sex with partners outside community and self‐reported genital ulcer disease in past year, sex and condom use with non‐stable partners)

^{^kk}

Number of HIV infections and py excluding one study for which information not presented (Auvert 2013 [33])