Miscarriage among women in the United States Women’s Interagency HIV Study, 1994-2017

Abstract

Background:

Relatively little is known about the frequency and factors associated with miscarriage among women living with HIV.

Objective:

To evaluate factors associated with miscarriage among women enrolled in the Women’s Interagency HIV Study (WIHS).

Study design:

We conducted an analysis of longitudinal data collected from October 1, 1994 to September 30, 2017. Women who attended at least two WIHS visits and reported pregnancy during follow-up were included. Miscarriage was defined as spontaneous loss of pregnancy before 20 weeks of gestation based upon self-report assessed at biannual visits. We modeled the association between demographic, behavioral, and clinical covariates and miscarriage (versus live birth) for women overall and stratified by HIV status using mixed model logistic regression.

Results:

Similar proportions of women living with and at without HIV experienced miscarriage (37% and 39%, respectively, p=0.638). In adjusted analyses, smoking tobacco (adjusted odds ratio, aOR=2.0), alcohol use (aOR=4.0), and marijuana use (aOR=2.0) were associated with miscarriage. Among women living with HIV, low HIV viral load (<4 log10 copies/mL) (aOR=0.5) and protease inhibitor (PI) (aOR=0.4) versus non-use of combination antiretroviral therapy (cART) use were protective against miscarriage.

Conclusions:

We did not find increased odds of miscarriage among women living with HIV compared to uninfected women, however poorly controlled HIV-infection was associated with increased miscarriage risk. Higher miscarriage risk among women exposed to tobacco, alcohol, and marijuana highlight potentially modifiable behaviors. Given previous concern about ART and adverse pregnancy outcomes, the novel protective association between PIs compared to non-cART and miscarriage in this study is reassuring.

INTRODUCTION

In the current combined antiretroviral treatment (cART) era, pregnancy rates have increased among US women with HIV and are comparable to women without HIV(1). Miscarriage occurs in an estimated 10-20% of US women(4-6). However, relatively little is known about the frequency and factors associated with miscarriage among women living with HIV (WLHIV).

Among the few cART era analyses, data are conflicting as to whether women with HIV experience an increased risk miscarriage; three studies from Africa reported increased risk(2-4) while one from the US did not(5). Most studies have focused on outcomes of low birth weight, preterm birth, neonatal intensive care unit admission, and neonatal morbidity(6-19), with fewer and more contradictory studies, mostly from sub-Saharan Africa, focused on miscarriage(2-5, 20-23).

Previous analyses from the Women’s Interagency HIV Study (WIHS), the largest ongoing multi-center prospective cohort study of HIV among women in the US, have explored factors associated with miscarriage. In the 1994-2002 WIHS cohort, Massad et al. found that miscarriage rates were similar for WLHIV and women without HIV, and that in WLHIV, miscarriage was associated with prior miscarriage, marijuana use, and non-ART use. That analysis did not consider the role of potential correlates including alcohol, frequency of marijuana use, sexually transmitted infections (STIs), specific cART regimens, or women without HIV(5). In the 1994-2012 WIHS cohort, Cates et al. found that HIV viral load during pregnancy predicted pregnancy loss(22), while Westreich et al. found cigarette smoking was associated with pregnancy loss in the 1994-2014 WIHS cohort(23). Neither of these studies reported on factors associated with miscarriage outside of viral load and smoking, respectively. To provide a more comprehensive and contemporary evaluation, we assessed the frequency and factors associated with miscarriage in both women living with and without HIV enrolled in WIHS.

MATERIALS AND METHODS

Cohort.

We analyzed data collected between October 1, 1994 and September 30, 2017 from women enrolled in WIHS(24). This cohort is comprised of women living with and without HIV recruited from ten US sites. In 1994, enrollment began in six sites (Bronx, Brooklyn, Chicago, Washington D.C., Los Angeles, San Francisco), and those sites re-enrolled during 2001-02 and 2011-12. The cohort was expanded in 2013-2015 to include four additional sites, to represent the southern United States (Atlanta, Chapel Hill, Miami, and Birmingham/Jackson). Enrollees were frequency matched on demographics and risk factors for acquiring HIV including drug use and number of sexual partners(24, 25). Medical and laboratory exams (including CD4 T-cell count and HIV viral load) as well as detailed interviews (collecting data including cART use, obstetric and gynecologic history, and medication use) are conducted longitudinally at semiannual visits by trained interviewers.

Ethics.

Each WIHS site and the data center provided Institutional Review Board approval prior to study enrollment. Enrollees provided written informed consent. All data used in this analysis were de-identified.

Outcome definition.

Women who attended at least two WIHS visits and self-reported a pregnancy during longitudinal follow-up were included in this analysis. Since prenatal care, delivery, and other obstetrical services were received outside of the study, pregnancy outcomes and dates were self-reported. Interviewers were trained to assess pregnancy and pregnancy outcome information from women by asking at every visit: “since your last study visit, how many times have you been pregnant?” and, for those responding with a non-zero number, “what was the outcome of the pregnancy?” Pregnancy outcomes were explained to help reduce misclassification error. Miscarriage was defined as spontaneous loss of a pregnancy before 20 weeks of gestation. Other pregnancy outcomes included stillbirth, defined as a child born dead after 20 weeks of gestation, live birth, defined as a child born alive, ectopic pregnancy, defined as any pregnancy outside the uterus, and abortion, defined as any pregnancy terminated through an elective procedure.

In total, 755 women reported 1487 pregnancies with known outcomes. We excluded pregnancies that ended in stillbirths (n=11, 0.7% of total pregnancies ending in stillbirth; n=5 women with only stillbirths) or ectopic pregnancies (n=47, 3% of total pregnancies ending in ectopic pregnancy; n=16 women with only ectopic pregnancies). We also excluded pregnancies without information in the visit prior to an outcome (n=64, 4% of total pregnancies; n=29 women) and those pregnancies that ended in abortion (n=454, 31% of total pregnancies ending in abortion; n=121 women with only abortion outcomes). We excluded women in the 2011/12 wave (n=11) and Southern sites (n=25) due to small numbers (comprising 44 pregnancies). In primary analyses, we compared women reporting miscarriage to the referent group of women reporting live births. In sensitivity analyses, the referent group also included abortions.

Covariates of interest.

Baseline covariates of interest measured at enrollment included: race/ethnicity, income, marital status, education, employment status, insurance status, WIHS site and enrollment wave, number of lifetime male sexual partners (categorized based on the distribution of the continuous variable and meaningful groupings), parity, prior miscarriage, and year of cART initiation for women with HIV. Due to the small number of self-reported prior STI events (<6 cases were recorded for each STI and none were individually associated with the outcome of interest), a composite variable for any prior STIs included gonorrhea, trichomonas, genital herpes, genital warts, syphilis, and chlamydia.

Time-varying covariates occur in the visit prior to the outcome of interest (and thus occur either during pregnancy or just prior to conception). Time-varying covariates of interest included: age group, alcoholic drinks per week (dichotomized as abstainers versus drinkers), marijuana use (dichotomized as yes/no) and frequency coded as none, less than weekly, 1-6 times a week, and at least daily), current cigarette smoking, any health related issues (including stroke, cancer, myocardial infarction, high blood pressure pulmonary embolism, diabetes), year of pregnancy, any new STI diagnosis (including gonorrhea, trichomonas, genital herpes, genital warts, syphilis, and chlamydia), yeast infection, bacterial vaginosis (BV) infection, and having a loop electrosurgical excision procedure.

Time-varying HIV-specific covariates of interest included: HIV viral load (dichotomized as ≥4 log₁₀ copies/mL of plasma versus <4 log₁₀ copies/mL of plasma for comparability with other WIHS analyses); CD4+ T-cell count (dichotomized as <350 cells/μL versus ≥350 cells/μL); ART regimen (categorized as (1) cART containing protease inhibitors (PI)-based regimen with nucleoside reverse transcriptase inhibitors (NRTI), (PI-based cART), (2) cART containing integrase inhibitors (II) and/or non-nucleoside reverse transcriptase inhibitors (NNRTI)-based regimens (non-PI cART) with NRTIs and (3) no cART including NRTI therapy alone or no ART); and ever having an AIDS defining illness.

Analysis methods.

We describe the frequencies of baseline categorical covariates using counts and percentages stratified by the primary outcome of interest (miscarriage versus live birth), with differences by miscarriage status quantified by Chi-square or Fisher’s exact tests, as appropriate. Frequencies for time-varying variables were calculated over all longitudinal visits. Descriptive analyses were calculated overall and stratified by HIV status.

To account for multiple pregnancy outcomes per woman, we modeled associations between covariates and miscarriage (versus live birth) for women overall and stratified by HIV status using mixed logistic regression models with a random intercept and an unstructured covariance matrix. Model-based point estimates (adjusted odds ratios, ORs) and 95% confidence intervals (CIs) are reported. Covariates significant in bivariate analyses at p<0.05 within a given strata (overall, living with or without HIV) were included in the corresponding multivariate models after assessing for multicollinearity. We decided a priori to include HIV status as a covariate in the model for women overall. Possible two-way interactions between marijuana use and cigarette smoking, marijuana use and drinking, and drinking and cigarette smoking were explored. In sensitivity analyses, models included both live births and abortion in the referent group. Analyses were performed with SAS v9.4 (SAS Institute, Cary, NC).

RESULTS

Miscarriage outcomes.

Among the 548 women included in the primary analysis, 207 (38%) experienced a miscarriage during follow-up and 341 (68%) only experienced live birth. Among 226 women without HIV, 88 (39%) ever experienced a miscarriage and 138 (61%) only experienced live birth. Among 322 WLHIV, 119 (37%) ever experienced a miscarriage and 203 (63%) only experienced live birth. There was no significant difference in the proportion of women ever having a miscarriage between WLHIV versus women without HIV (37% versus 39%, respectively, p=0.638).

Baseline descriptive findings (Table 1).

Table 1.

Baseline characteristics stratified by HIV status and miscarriage, WIHS 1994-2017

		All women (N = 548)					Women living with HIV (N= 322)					Women without HIV (N = 226)
		Mis carriage (n=207, 38%)		Live birth (n=341, 66%)		p- value	Mis carriage (n=119, 37%)		Live birth (n=203, 63%)		p- value	Mis carriage (n=88, 39%)		Live birth (n=138, 61%)		p- value
Variables		n	%	n	%		n	%	n	%		n	%	n	%
Race/ethnicity	White/Other	22	11	48	14	0.193	9	8	29	14	0.140	13	15	19	14	0.494
	African American	130	63	188	55		79	66	117	58		51	58	71	51
	Hispanic	55	27	105	31		31	26	57	28		24	27	48	35
Income/year	≤$12,000	108	55	184	56	0.951	62	54	109	56	0.726	46	55	75	56	0.539
	$12-24,000	47	24	75	23		27	24	51	26		20	24	24	18
	>$24,000	43	22	70	21		25	22	36	18		18	21	34	26
Marital status	Legally/common-law married	28	14	67	20	0.082	17	14	42	21	0.117	11	13	25	18	0.525
	Unmarried but living with partner	36	17	70	21		21	18	47	23		15	17	23	17
	Single/Widowed	142	69	204	60		80	68	114	56		62	70	90	65
Education	Less than high school	125	61	200	59	0.668	72	61	115	57	0.499	53	61	85	62	0.866
	High school or higher	81	39	140	41		47	39	88	43		34	39	52	38
Employed	No	147	71	237	70	0.708	90	76	150	74	0.730	57	65	87	63	0.792
	Yes	60	29	104	30		29	24	53	26		31	35	51	37
Insurance	No	48	23	108	32	0.035	18	15	51	25	0.041	30	34	57	42	0.241
	Yes	157	77	231	68		99	85	152	75		58	66	79	58
Site group	New York	105	51	130	38	0.038	62	52	76	37	0.071	43	49	54	39	0.281
	District of Columbia	27	13	53	16		15	13	38	19		12	14	15	11
	California	53	26	111	33		27	23	61	30		26	30	50	36
	Chicago	22	11	47	14		15	13	28	14		7	8	19	14
WIHS enrollment wave	Original cohort (1994/95)	106	51	149	44	0.087	69	58	98	48	0.701	37	42	51	37	0.444
	2001/02 Recruitment	101	49	192	56		50	42	105	52		51	58	87	63
Number lifetime male partners	0-4	43	21	103	31	0.013	24	21	73	37	0.001	19	22	30	22	0.978
	5-24	94	46	155	46		51	44	86	43		43	49	69	50
	>=25	67	33	78	23		41	35	39	20		26	30	39	28
Parity	0	62	30	89	26	0.313	27	23	48	24	0.530	35	40	41	30	0.268
	1-2	84	41	161	47		51	43	97	48		33	38	64	46
	>=3	61	29	91	27		41	34	58	29		20	23	33	24
Prior miscarriage	No	130	63	237	70	0.106	71	60	149	73	0.011	59	67	88	64	0.614
	Yes	77	37	104	30		48	40	54	27		29	33	50	36
Prior STI*	No	82	40	137	40	0.853	39	33	78	39	0.279	43	49	59	43	0.368
	Yes	125	60	202	60		80	67	123	61		45	51	79	57
Year of cART initiation	1995-99						61	54	98	51	0.092
	2000-3						32	29	61	32
	2004-8						15	13	21	11
	2009-						4	4	11	6

WIHS: Women's Interagency HIV Study; STI: sexually transmitted infection; cART: combination highly active antiretroviral therapy

^*Includes gonorrhea, trichomonas, genital herpes, genital warts, syphilis, chlamydia

p-values are two-tailed

The majority of women were African American (58%), earned <=$12,000 USD per year (55%), had less than a high school education (60%), were unemployed (70%), and had between 1-2 children (45%) upon entry into WIHS. Among all women, those experiencing a miscarriage were more likely (p<0.05) to be insured, be recruited from the New York WIHS sites, and have a higher number of lifetime male sex partners. Among WLHIV, those experiencing a miscarriage were more likely (p<0.05) to be insured, have a higher number of male sex partners, and have had a prior miscarriage. No differences were observed between baseline covariates and miscarriage in women without HIV.

Time-varying descriptive findings (Table 2).

Table 2.

Time-varying characteristics stratified by HIV status and miscarriage, WIHS 1994-2017

		All women (N = 848)					Women living with HIV (N = 468)					Women without HIV (N = 380)
		Mis carriage (n=250, 29%)		Live birth (n=598, 71%)		p-value	Mis carriage (n=140, 30%)		Live birth (n=328, 70%)		p-value	Mis carriage (n=110, 29%)		Live birth (n=270, 71%)		p-value
		n	%	n	%		n	%	n	%		n	%	n	%
Variables (previous visit)
Age group	<25	63	25	97	16	0.011	29	21	38	12	0.062	34	31	59	22	0.138
	25-30	50	20	155	26		29	21	90	27		21	19	65	24
	30-35	65	26	189	32		41	29	105	32		24	22	84	31
	>35	72	29	157	26		41	29	95	29		31	28	62	23
Alcohol use	Abstain	118	49	450	81	<.0001	76	56	263	84	<.0001	42	40	187	77	<.0001
	Drinkers	123	51	108	19		60	44	51	16		63	60	57	23
Alcohol use frequency	Abstain	118	49	450	81	<.0001	76	56	263	84	<.0001	42	40	187	77	<.0001
	0-7	101	42	89	16		46	34	45	14		55	52	44	18
	>7	22	9	19	3		14	10	6	2		8	8	13	5
Marijuana use	No	171	70	492	88	<.0001	105	77	288	91	<.0001	66	62	204	83	<.0001
	Yes	72	30	68	12		32	23	27	9		40	38	41	17
Marijuana use frequency	None	171	71	492	88	<.0001	105	77	288	91	0.001	66	64	204	84	0.001
	Less than weekly	15	6	24	4		6	4	11	3		9	9	13	5
	1-6 times/week	22	9	16	3		12	9	7	2		10	10	9	4
	At least daily	32	13	27	5		14	10	9	3		18	17	18	7
Cigarette smoking	No	120	49	401	72	<.0001	65	47	231	73	<.0001	55	52	170	69	0.003
	Yes	123	51	159	28		72	53	84	27		51	48	75	31
Health related*	No	193	77	477	80	0.414	108	77	256	78	0.829	85	77	221	82	0.334
	Yes	57	23	121	20		32	23	72	22		25	23	49	18
Pregnancy year	1994-1997	50	20	86	14	0.066	36	26	49	15	0.023	14	13	37	14	0.655
	1998-2001	46	18	85	14		29	21	57	18		17	16	28	10
	2002-5	74	30	215	36		39	28	124	38		35	32	91	34
	≥2006	79	32	208	35		36	26	95	29		43	39	113	42
Any STI diagnosis**	No	211	86	493	88	0.564	114	83	265	84	0.688	97	92	228	93	0.608
	Yes	33	14	67	12		24	17	50	16		9	8	17	7
Yeast infection	No	200	82	434	78	0.120	111	81	237	75	0.177	89	84	197	80	0.509
	Yes	43	18	126	23		26	19	78	25		17	16	48	20
BV infection	No	186	80	356	85	0.134	97	73	192	81	0.066	89	90	164	90	0.873
	Yes	46	20	63	15		36	27	44	19		10	10	19	10
LEEP procedure	No	248	99	596	100	0.389	138	99	327	100	0.208	110	100	269	100	n/a
	Yes	2	1	2	0		2	1	1	0		0	0	1	0
HIV viral load	≥4 log10 copies/mL						45	33	43	14	<.0001
	<4 log10 copies/mL						90	67	266	86
CD4 count <500 cells/μL	No						57	43	127	41	0.759
	Yes						77	57	183	59
cART regimen	II, NRTI, NNRTI						23	17	71	22	<.0001
	PI						28	20	128	40
	No cART						87	63	121	38
AIDS defining illness	No						98	70	247	75	0.235
	Yes						42	30	81	25

WIHS: Women's Interagency HIV Study; cART: combination highly active antiretroviral therapy; II: integrase inhibitor; NNRTI: non-nucleoside reverse transcriptase inhibitors; NRTI: nucleoside reverse transcriptase inhibitors; PI: protease inhibitor; STI: sexually transmitted infection; BV: bacterial vaginosis; LEEP: loop electrosurgical excision procedure

^*Includes stroke, cancer, myocardial infarction, high blood pressure pulmonary embolism, diabetes

^**Includes gonorrhea, trichomonas, herpes-2, genital warts, syphilis, chlamydia

‘n’ is given over all study intervals

p-values are two-tailed

In the visit prior to pregnancy outcomes of interest, the majority of women were older than 30 (57%), did not use alcohol (71%) or marijuana (83%), did not smoke (65%), and had no STI diagnosis (88%). Among all women, those experiencing a miscarriage were more likely (p<0.05) to be alcohol drinkers, marijuana users, and cigarette smokers. WLHIV experiencing a miscarriage were also more likely (p<0.05) to have experienced a pregnancy before 1998, have a higher viral load, and not be taking cART.

Adjusted model: All women (Table 3).

Table 3. Baseline and time-varying factors associated with miscarriage in all women, WIHS 1994-2017.

	Level	aOR	95% CI		p-value
HIV status	HIV+ (ref: HIV−)	1.23	0.84	1.79	0.280
Insurance (baseline)	Yes (ref: No)	1.11	0.72	1.71	0.628
Site group	New York (ref: California)	1.84	1.18	2.86	0.007
	District of Columbia (ref: California)	1.09	0.58	2.04	0.789
	Chicago (ref: California)	0.80	0.41	1.56	0.502
Number male partners	5-24 (ref: 0-4)	0.99	0.63	1.55	0.962
	>=25 (ref: 0-4)	1.21	0.72	2.03	0.471
Age group (previous visit)	25-30 (ref: <25)	0.48	0.28	0.82	0.007
	30-35 (ref: <25)	0.56	0.34	0.92	0.022
	>35 (ref: >25)	0.63	0.38	1.05	0.078
Alcohol use (previous visit)	Drinkers (ref: abstainers)	3.97	2.71	5.81	<.0001
Marijuana use (previous visit)	Yes (ref: No)	1.68	1.06	2.67	0.029
Cigarette smoking (previous visit)	Yes (ref: No)	1.98	1.35	2.91	0.001

Mixed models with random intercept and unstructured covariance matrix

aOR: adjusted odds ratio; CI: confidence interval; ref: reference; WIHS: Women's Interagency HIV Study

p-values are two-tailed

Factors associated (p<0.05) with miscarriage (versus live birth) included: being enrolled in the New York WIHS sites versus California sites, drinking alcohol, using marijuana in the previous visit, and cigarette smoking. Older age was protective for miscarriage. HIV status was not associated with miscarriage (p=0.280). None of the two-way interactions between marijuana use and cigarette smoking, marijuana use and drinking, and drinking and cigarette smoking were significant. In sensitivity analyses when elective abortions were included in the referent group, we found similar adjusted findings (data not tabled).

Adjusted model: WLHIV (Table 4).

Table 4.

Baseline and time-varying factors associated with miscarriage in women living with HIV, WIHS 1994-2017

Variable	Level	aOR	95% CI		p-value
Insurance (baseline)	Yes (ref: No)	2.30	1.12	4.72	0.024
Number male partners	5-24 (ref: 0-4)	1.05	0.55	1.99	0.885
	>=25 (ref: 0-4)	1.02	0.49	2.14	0.955
Prior miscarriage	Yes (ref: No)	1.25	0.73	2.14	0.412
Alcohol use (previous visit)	Drinkers (ref: abstainers)	2.96	1.71	5.12	<0.001
Marijuana use (previous visit)	Yes (ref: No)	1.58	0.75	3.32	0.226
Cigarette smoking (previous visit)	Yes (ref: No)	2.20	1.26	3.85	0.006
Pregnancy year	1998-2001 (ref: 1994-1997)	1.02	0.45	2.34	0.961
	2002-2005 (ref: 1994-1997)	0.80	0.37	1.75	0.576
	>=2006 (ref: 1994-1997)	1.30	0.56	3.05	0.540
HIV viral load <4 log10 copies/mL (previous visit)	Yes (ref: No)	0.45	0.24	0.84	0.013
ART regimen	cART containing II, NRTI, NNRTI (ref: No cART)	0.69	0.34	1.40	0.302
	cART containing PI (ref: No cART)	0.40	0.21	0.79	0.008

Mixed models with random intercept and unstructured covariance matrix

aOR: adjusted odds ratio; CI: confidence interval; ref: reference; WIHS: Women's Interagency HIV Study; cART: combination highly active antiretroviral therapy; NNRTI: non-nucleoside reverse transcriptase inhibitors; NRTI: nucleoside reverse transcriptase inhibitors; PI: protease inhibitor

p-values are two-tailed

Factors associated (p<0.05) with miscarriage (versus live birth) included: having insurance, drinking alcohol, and cigarette smoking. Lower log HIV viral load (<4 log10 copies/mL) and using a PI regimen (versus no cART) were protective for miscarriage. Women with HIV viral load <4 log10 copies/mL experienced similar miscarriage proportions compared to women without HIV (data not tabled). None of the two-way interactions between marijuana use and cigarette smoking, marijuana use and drinking, and drinking and cigarette smoking were significant. In sensitivity analyses when abortion was included in the referent group, we found similar adjusted findings.

Adjusted model: Women without HIV (Table 5).

Table 5.

Baseline and time-varying factors associated with miscarriage in women without HIV, WIHS 1994-2017

Variable	Level	aOR	95% CI		p-value
Alcohol use (previous visit)	Drinkers (ref: abstainers)	4.16	2.38	7.28	<.0001
Marijuana use (previous visit)	Yes (ref: No)	1.82	0.97	3.43	0.062*
Cigarette smoking (previous visit)	Yes (ref: No)	1.77	1.01	3.10	0.047

Mixed models with random intercept and unstructured covariance matrix

aOR: adjusted odds ratio; CI: confidence interval; ref: reference; WIHS: Women's Interagency HIV Study

^*Point estimate is significant in sensitivity analyses including abortion in reference group (aOR=1.71; 95%CI: 1.02-2.86, p=0.041)

p-values are two-tailed

Factors associated (p<0.05) with miscarriage (versus live birth) included: drinking alcohol and cigarette smoking. None of the two-way interactions between marijuana use and cigarette smoking, marijuana use and drinking, and drinking and cigarette smoking were significant. In sensitivity analyses when abortion was included in the referent group, we found similar adjusted findings although marijuana use became statistically significant (aOR=1.71; 95%CI: 1.02-2.86, p=0.041).

Marijuana frequencies (Figure 1).

Figure 1.

Adjusted association between marijuana use frequency categories and miscarriage, WIHS, 1994-2017

Odds ratios adjusted for same covariates as primary analyses

WIHS: Women's Interagency HIV Study; CI: confidence interval

*p=0.0320 comparing 'At least daily' versus 'None'

**p=0.0324 comparting 'At least daily' versus 'None to less than weekly'

In addition to dichotomized marijuana use (yes/no), frequency of marijuana use was also significant in bivariate analyses. Among all women, only the highest category of marijuana use (at least daily) reached statistical significance (p=0.032) compared to either no marijuana use or none/less than weekly marijuana use. Similar findings were observed for women living with and without HIV, though the adjusted point estimates for frequency of marijuana use did not reach statistical significance after stratifying by HIV status.

STRUCTURED DISCUSSION/COMMENT

1. Principal Findings.

In this longitudinal study, a similar proportion of women living with and without HIV reported miscarriage (37% and 39%, respectively; p=0.638). These proportions are similar to previous (1994-2002) WIHS analyses which observed 41% of women living with and 34% of women without HIV reporting miscarriage (among women reporting either a miscarriage or live birth)(5). These proportions are considerably higher than the general population in which miscarriage occurs in an estimated 10-20% of US women(4-6), which was not surprising given that women in the WIHS are on average of lower socio-economic status, report more substance use, and have more genital infections relative to the general population, all of which have been associated with miscarriage in other studies(18, 26-31).

Our study adds several years of follow-up to confirm previous WIHS findings including no association between miscarriage and HIV status(5), a positive association between miscarriage and higher HIV viral load(22), and an effect for cigarette use that is stronger among women living with versus without HIV(23). We also identified new factors associated with miscarriage: alcohol use, marijuana use frequency, and a protective effect for PI versus non-cART use. Interestingly, history of miscarriage was not associated with miscarriage.

2. Results.

Alcohol use was associated with miscarriage in this analysis. Heavy alcohol use in pregnancy has been related to negative birth outcomes, including miscarriage, and the amount and pattern of drinking likely influences these outcomes(26-28). Though we did not have sufficient sample size to evaluate moderate and heavy drinking, a systematic review reported that low to moderate drinking is not associated with negative birth outcomes including miscarriage(32). However, given that low to moderate alcohol consumption has not been shown to be definitively safe during pregnancy, US medical societies and the CDC recommend alcohol avoidance when trying to conceive or during pregnancy/breastfeeding(33).

In a previous (1994-2002) WIHS analysis by Massad et al, marijuana use in two visits prior to pregnancy in women with HIV was associated with miscarriage (aOR=6.6 compared to women with no or inconsistent use), but findings for women without HIV were not explored, nor was the effect of frequency of marijuana use for women living with and without HIV(5). In the present analysis, we found that any marijuana use was associated with miscarriage overall with a trend for increasing marijuana use. Though publications have demonstrated a potential biological mechanism for the effect of marijuana on miscarriage (primarily related to the harmful effect of delta-9-tetrahydrocannabinol, the main psychoactive in marijuana which crosses the placental barrier)(34-36), several studies failed to show an association between marijuana use and miscarriage. Earlier (1987-2002) longitudinal studies of prenatal marijuana exposure in women without HIV found no association between marijuana use during pregnancy and miscarriage(7, 37-40). However, contemporary marijuana products have higher levels of delta-9-tetrahydrocannabinol than in the 1980-90s(41). As access to marijuana increases through legalization and stigmatization diminishes, women may be more likely to accurately disclose their marijuana use and thus future studies may be more accurate. Given conflicting literature on marijuana use and adverse birth outcomes including miscarriage, low birth weight, preterm birth, and neonatal morbidity(6-12, 18, 37-40, 42), professional societies recommend women discontinue marijuana during pregnancy, while breastfeeding, or when attempting to conceive(43).

In a previous WIHS analysis, ART at the visit prior to pregnancy appeared protective against miscarriage (aOR=0.37)(5). Here, we explored specific cART regimens and found that PI-containing regimens versus no cART were significantly protective for miscarriage, while other cART regimens may also be protective (though their association did not reach statistical significance). While data are conflicting, some studies show cART to be associated with adverse pregnancy outcomes including preterm delivery and low birth weight(13-15). A review of studies published between 1980-2017 compared women receiving tenofovir versus non-tenofovir cART regimens during pregnancy and found no increased miscarriage risk(44), and a study of zidovudine versus non-zidovudine cART regimens found a protective effect for miscarriage(45). Meanwhile, PIs may lower progesterone levels(46) possibly leading to adverse pregnancy outcomes including miscarriage(47), preterm birth(16), and lower birth weight(17).

History of miscarriage was not associated with miscarriage in our adjusted models. This finding contradicts Massad et al in which miscarriage was significantly associated with prior miscarriage in women with HIV (aOR=1.94)(5). This difference may be related to our inclusion of an additional 15 years of study data and control for influential covariates such as alcohol and cART. Importantly, we are not differentiating between 1-2 versus recurrent prior miscarriage (3 or more), the latter being much less common and likely reflective of underlying factors that may increase miscarriage risk.

3. Research Implications.

The protective association with PIs persists when controlling for viral load, indicating that possible mechanism of protection is not conferred by viral suppression. This mechanism warrants further exploration.

4. Clinical Implications.

Alcohol consumption and marijuana use are potentially modifiable behaviors that providers can counsel on to impact miscarriage risk. The finding that PIs appeared significantly protective for miscarriage importantly adds to PI safety data.

5. Strengths and Limitations.

Several limitations warrant discussion. First, miscarriage in the WIHS is self-reported. The risk of miscarriage is highest in the first several weeks of pregnancy(48) when women and their providers may not diagnose early pregnancy(49, 50). This could contribute to ascertainment bias. Additionally, use of marijuana, alcohol, tobacco use, and STI could be differentially reported by the outcome of interest, with women experiencing miscarriage perhaps less likely to report exposure. However, self-reported data in other WIHS studies has been strongly correlated with clinical outcomes. We did not have sufficient numbers to explore the role of individual STIs or levels of drinking. We also did not have measures of time-varying intimate partner violence, which may be associated with miscarriage. Future analyses could consider the role of various measures of smoking frequency and different time-varying combinations of ART regimens. Though rare, it is possible that a proportion of miscarriages could include early ectopic pregnancies. Though we excluded the 2011/2012 wave and southern due to small numbers, when combined with the analysis cohort, findings did not meaningfully change. As the southern cohort accrues more women, future analyses will be able to draw conclusions from this group. Finally, our results are most generalizable to US urban women living with and without HIV who are primarily older and African American.

6. Conclusions.

The WIHS is one of the largest and longest running cohorts of women living with and without HIV in the world and is well-validated with standardized data collection procedures and highly trained staff. Our study adds several years of analytic follow-up to confirm previous findings and identify new findings related to alcohol use, marijuana use frequency, and the protective effect of PIs. These findings highlight additional potentially modifiable behaviors addressable via interventions which could reduce miscarriage risk.

CONDENSATION:

Poorly controlled HIV-infection and tobacco, alcohol, and marijuana use were associated with increased miscarriage risk while protease inhibitors were protective compared to non-combination antiretroviral treatment.

AJOG AT A GLANCE:

A. To provide a more comprehensive and contemporary evaluation of miscarriage, we assessed the frequency and factors associated with miscarriage in both women living with and without HIV enrolled in the Women's Interagency HIV Study (WIHS).

B. HIV status was not associated with miscarriage risk. Poorly controlled HIV-infection and tobacco, alcohol, and marijuana use were associated with increased miscarriage risk while protease inhibitors were protective compared to non-combination antiretroviral treatment.

C. Our study adds several years of analytic follow-up to previous analyses, confirms previous findings related to HIV status and miscarriage, and identified new findings related to alcohol use, marijuana use frequency, and the protective effect of protease inhibitors.