Racial/Ethnic Inequities in Paid Parental Leave Access

Purpose: Severe racial inequities in maternal and infant health in the United States are caused by the many forms of systemic racism. One manifestation of systemic racism that has received little attention is access to paid parental leave. The aim of this article is to characterize racial/ethnic inequities in access to paid leave after the birth of a child. Methods: We analyzed data on women who were employed during pregnancy (n=908) from the Bay Area Parental Leave Study of Mothers, a survey of mothers who gave birth in the San Francisco Bay Area in 2016–2017. We examined differences in access to government- and employer-paid leave, the duration of leave taken, and the percent of usual pay received while on leave. To explore these differences, we further examined knowledge of paid leave benefits and sources of information. Results: Non-Hispanic (NH) black and Hispanic women had significantly less access to paid leave through their employers or through government programs than their NH white and Asian counterparts. Relative to white women, Asian, Hispanic, and black women received 0.9 (p<0.05), 2.0 (p<0.01), and 3.6 (p<0.01) fewer weeks, respectively, of full-pay equivalent pay during their parental leaves. Despite inequitable access to paid leave, the duration of parental leave taken did not differ by race/ethnicity. Conclusions: Inequitable access to paid parental leave through both employers and government programs exacerbates racial inequities at birth. This form of structural racism could be addressed by policies expanding access to paid leave.


Introduction
Striking racial inequities in maternal and infant health in the United States are well documented. The rate of severe maternal morbidity, or life-threatening pregnancy complications, has increased significantly over recent decades for all racial/ethnic groups, but remains consistently higher for non-Hispanic (NH) black, Hispanic, and Asian/Pacific Islander women compared to NH white women, even after adjusting for covariates. 1 NH black women, in particular, are more than three times as likely to die from pregnancy-related causes than NH white women 2 and face significantly higher risks of preterm birth and other adverse birth outcomes. 3 The infant mortality rate among NH black infants is more than twice as high as that among NH white infants. 4 Black infants are significantly less likely to initiate breastfeeding and, among those who do, are less likely to be breastfeed for at least 6 months. 5 These inequities are thought to be caused by numerous factors, including systemic and interpersonal racism. 6,7 Structural racism, which operates at the macrolevel and influences systems, ideologies, social forces, and institutions, has been identified as the root cause of a range of health inequities. 8 Allostatic load, or the cumulative physiological effects of stress over the life course, has gained traction as a theory to explain racial inequities in pregnancy outcomes due to the burden of racism experienced by women of color. 9 In addition, structural racism in health care leads to black women often receiving poorer quality care relative to white women. For example, residential segregation rooted in racism leads to black women being more likely to deliver at poor quality hospitals 10 and black patients' reports of pain being taken less seriously than those of white patients. 11 Women who experience perceived discrimination during prenatal visits, including discrimination based on their race or ethnicity, have more than twice the odds of nonattendance at postpartum visits, 12 which can contribute to increased risk for maternal mortality.
Another manifestation of structural racism that has received less attention is differential access to paid parental leave. Substantial research documents beneficial effects of parental leave-taking for both new parents and their children. [13][14][15][16][17][18][19][20][21][22] Paid parental leave, which enables workers to take time away from work to recover from childbirth and care for a new baby, is associated with decreased low-birth-weight births and infant mortality, increased breastfeeding, and improved maternal mental health. [19][20][21][22][23][24][25][26] In 2018, the American College of Obstetricians and Gynecologists (ACOG) endorsed at least 6 weeks of fully paid leave for all new mothers to reduce high rates of maternal mortality in the United States. 27,28 Paid leave confers economic as well as health benefits. Availability of paid leave increases household income, decreases risk of poverty, and reduces some forms of material hardship, especially among lesseducated and low-income single mothers. 29,30 (This is in addition to the fact that access to paid leave itself is also a marker of the underlying racial segregation of labor markets that also directly contribute to racial differences in wages, as discussed further below.) Importantly, these health and labor market benefits are largely associated with paid, but not unpaid, leave. 16 The small body of work examining racial/ethnic differences in the impact of paid leave policies on economic and health outcomes has found mixed results. For example, Rossin-Slater et al. found that, while California's paid family leave (CA-PFL) program increased leave-taking for all mothers, the largest absolute gain in leave-taking was among black mothers who, as a group, had the lowest baseline levels of maternity leave-taking. 31 This is reinforced by work from Joshi and colleagues showing that black and Hispanic parents are less likely than their white counterparts to be eligible for and potentially able to afford unpaid leave. 32 Other reports show similar eligibility for unpaid leave across racial groups, yet higher unmet need for leave among black workers relative to white and Asian workers, suggesting that eligibility for jobprotected leave alone potentially masks financial or other limitations on workers' ability to take leave. 33 Simulations of leave-taking under various policy scenarios suggest that paid family and medical leave (PFML) (vs. unpaid leave) would enable black workers to gain a greater percentage of their family income back relative to white workers, but would have a relatively smaller impact on preventing short-term economic hardship. 34 In terms of health outcomes, Hamad et al. examined paid family leave policies in California and New Jersey and found that, relative to white women, black women had reduced breastfeeding at 12 months, while Hispanic women had increased exclusive breastfeeding at 6 months. 20 Recent studies of infant immunizations 35 and birth outcomes 36 found no racial differences, while others report that small subgroup sample sizes prevented detection or examination of racial/ethnic differences. 22,24 The lack of a national PFML program and, instead, reliance on a patchwork of state and local policies that provide partial pay for some workers mean that many of the most economically vulnerable parents lack access to paid leave. The Family and Medical Leave Act provides job-protected, unpaid caregiving leave to covered workers, but strict eligibility requirements exclude about half of the workforce. 37 Access to paid leave, which financially enables many lower income workers to take advantage of leave benefits, is even more limited: just eight states (CA, RI, NJ, NY, WA, MA, CT, and OR) and the District of Columbia have government PFML programs. 38 Only about 19% of U.S. workers have any paid family leave offered through their employers, with even lower access among part-time workers, nonunionized workers, workers in low-wage industries, and workers at small firms. 39 Surprisingly, little research has examined racial and ethnic inequities in access to paid leave. These inequities could stem from occupational segregation affecting the likelihood of working for a firm that offers paid leave and job characteristics, such as working full versus part time and job tenure, which could impact eligibility for employer-and government-paid leave benefits.
Two analyses draw on data from a 2011 Leave Module in the Bureau of Labor Statistics' American Time Use Survey that asked whether respondents received any paid leave from their employer and whether they were able to take paid leave for the birth or adoption of a child, for their own medical condition, to care for a family member, or for other reasons. These analyses show that just 23-25% of Hispanic parents have access to paid parental leave, compared to 47-50% of NH white and 41-43% of NH black parents. 40,41 This unequal access to paid parental leave does not appear to be due to differences in the likelihood of working full time: Hispanic workers are equally likely to be employed full versus part time compared to NH workers. 40 Differential access more likely stems from the types of jobs workers hold. Hispanic workers are most likely to fall in the lowest wage brackets 40 and, relative to white and Asian workers, both NH black and Hispanic workers are underrepresented in professional-class jobs 42 -jobs that are more likely to provide benefits like paid leave. 40 Once demographic and employment characteristics are controlled for, the differences in access to paid parental leave between NH whites and blacks go away; however, Hispanic workers remain significantly less likely to receive these benefits. 41 Decomposition of these differences in paid leave access points to the importance of immigration and citizenship status, as well as differential returns to full-time employment for Hispanic workers. 41 This also suggests additional mechanisms, such as potentially a lack of awareness of available benefits or lack of adequate support from employers to facilitate leave-taking. In a survey of California workers who had recently become a parent or had a close family member become seriously ill, less than half of respondents were aware of the state's Paid Family Leave program, and awareness was lowest among low-wage workers (38%), immigrants (34%), Latinx workers (34%), and workers with less than a high school diploma (21%). 43 Workers of color may also be less likely to work in jobs that are covered by PFML policies, where applicable. For example, most public sector (i.e., governmental) workers are excluded from paid family leave programs in California, Rhode Island, New York, District of Columbia, and Connecticut 38 (although some states, including California, allow public sector workers to opt in) and black parents are overrepresented in public sector jobs. 42 In addition, all state PFML pro-grams have some employee eligibility requirements that specify minimum earnings and/or hours worked in the base period, which may disproportionately exclude those in part-time, seasonal, or low-wage jobsjobs in which women of color are overrepresented. 33,42 Importantly, the amount of leave actually taken after giving birth does not appear to vary dramatically across racial/ethnic groups, and recent studies have shown inconsistent leave-taking patterns. 41,[44][45][46] If parents take the same amount of leave regardless of pay, this implies that non-white parents are less likely to receive financial support in the newborn period-parents who already own far less wealth than white parents as the result of racist social and economic policies, racial segregation, and slavery. 47 Even among workers who receive some paid leave, the amount and duration of pay may vary. While employer-provided paid leave often provides full pay for a set number of weeks, state PFML programs offer a percentage of usual wages. 38 The consequence is that workers relying exclusively on state PFML programs may face a more severe financial penalty for taking leave relative to workers receiving fully paid leave through their employers. The extent to which this impacts workers, however, remains unknown as most existing data sources do not differentiate between government-and employer-provided pay, nor do they assess the proportion of usual pay received while on leave. This gap in the literature leaves the full impact of inequitable access to paid leave unknown.
This study draws upon a unique data source that includes detailed questions on the amount and source(s) of pay available after the birth of a child to better understand racial/ethnic inequities in paid leave access.
The objective of our study was to examine whether there are racial/ethnic inequities in (1) access to government-and employer-paid leave; (2) amount of pay received while on leave; and (3) the duration of leave taken after the birth of a child. To better understand these inequities, we also examined differences in knowledge of paid leave benefits and sources of information. Finally, to summarize the magnitude of the inequity in paid leave access, we calculated racial/ ethnic disparities in the average number of full-pay equivalent (FPE) weeks of paid leave taken.
We present all our results without controlling for covariates to avoid concealing the influence of important sociodemographic and job characteristics that may lead to differential access and utilization of paid leave benefits.

Data
We use data from the Bay Area Parental Leave Study of Mothers, a survey of mothers who gave birth in the San Francisco Bay Area in 2016-2017. We mailed invitations to a stratified random sample from birth records for all women who delivered live births in the San Francisco Bay Area in 2016 or 2017 to participate in a 25min online survey. Women who were employed or whose partners were employed during pregnancy, whose child was still living with them, and who could complete the survey in English or Spanish were eligible to participate. Participants were offered a $15 gift card for their participation. Nonresponders were mailed up to two reminder cards (the second with a higher incentive offer); the subset of remaining nonrespondents with identifiable phone numbers was subsequently called.
We completed data collection for 2016 births in August 2018 and for 2017 births in January 2019. One thousand three hundred-four women completed our survey (20.4% response rate). Our analytic sample for this article includes respondents who were employed by someone else during pregnancy (i.e., excluding nonemployed and self-employed women) and whose selfreported race was NH white (hereafter referred to as white), NH Asian (hereafter referred to as Asian), NH black (hereafter referred to as black), or Hispanic of any race (hereafter referred to as Hispanic) (N = 908).
The Bay Area Parental Leave Study of Mothers was originally conducted to examine the impact of the San Francisco Paid Parental Leave Ordinance (PPLO), a paid leave policy that went into effect for San Francisco employers in 2017, on leave-taking among new mothers. Previous analyses showed that the PPLO had no impact on leave-taking among women in the Bay Area. 48 Thus, for this analysis of access to paid leave, we combine pre-and post-PPLO waves of data and all employed women, regardless of county of employment.

Measures
Our primary independent variable is mothers' selfreported race/ethnicity. We restrict to women who identified as white, Asian, black, or Hispanic (we drop 6.6% of our sample who reported their race as ''other'' or who had missing race/ethnicity).
Employed women were asked a series of questions about their job(s) during pregnancy and whether and how much leave from work they took after the birth of their child. All respondents took some postnatal parental leave, so our analyses focus on duration only. We assessed the duration of postnatal parental leave among women who returned or planned to return to their same job with the question, ''How old was your child when you returned to working for pay, even part time?'' Women who had not yet returned to work, but who planned to were categorized as taking 12 or more weeks of leave, as all interviews took place greater than 12 weeks postpartum.
We assessed paid leave access using four measures: (1) we asked women the following question: ''Thinking about all sources of pay during your leave, was the amount of money you received equal to your regular pay?'' Women who said no were asked to report the percent of total pay on average they received while on leave. Combining these two questions, we categorized the percent of usual pay as no pay, less than 50% of regular pay, 50-99% of regular pay, and 100% of regular pay.
(2) We categorized employer-provided paid leave based on whether the respondent reported that their employer provided paid leave and, if so, for how many weeks: no paid leave, 1 to 5 weeks, 6 to 11 weeks, and 12 or more weeks.
(3) We examined government-provided paid leave, which, in California, takes the form of State Disability Insurance and Paid Family Leave and together provide 12 weeks (14 weeks for Cesarean delivery) of paid leave for parents recovering from childbirth. We therefore made the assumption that women who reported receiving any government-provided paid leave and who took at least 12 weeks (14 for Cesarean delivery) of leave received paid leave from the government for 12 or 14 weeks. For women who reported receiving governmentprovided paid leave and who took less than 12 weeks of leave, we assumed that they received paid leave from the government for the duration of their leave. We therefore categorized government-paid leave as no paid leave, 1 to 11 weeks, or 12 or more weeks. Note that these programs do not cover public sector (i.e., government) employees unless their employer opted in. We therefore do not know whether individual employees, particularly those who are employed in the public sector, are eligible for paid leave through these programs.
(4) We multiplied the percent of usual pay during leave by the number of weeks of leave taken to estimate the FPE weeks of paid leave. For respondents whose percent of usual pay was provided as a range, we used the midpoint (e.g., ''less than 50%'' was assigned 25% and ''51-75%'' was assigned 63%). As a sensitivity analysis, we assigned those reporting ''less than 50%'' first no pay and then 49% pay.
Additional outcomes of interest included whether respondents understood the maternity leave benefits available to them and whether their employer or anyone else was a source of information about their maternity leave benefits.
We present demographic and job characteristics by race/ethnicity to inform interpretation of our results. Job characteristics include public versus private sector, job tenure, hours worked during pregnancy, employer size, and employer location; sociodemographic characteristics include marital/partner status, education, language spoken at home, and Medicaid coverage during pregnancy as a proxy for household income.

Approach
We used unadjusted linear probability models and ordinary least squares models to examine differences in leave-taking probability and duration across white, black, Asian, and Hispanic women. We reported robust standard errors and incorporated probability weights to account for nonresponse and oversampling of San Francisco residents, Spanish speakers (proxied by mother's immigration from a Spanish-speaking country), and low-income (proxied by Medicaid status) women.
Racial/ethnic disparities in access to paid leave could result from differences in ''downstream'' characteristicsdemographic or human capital differences and geographic and occupational segregation-that are produced by ''upstream'' processes of discrimination (e.g., inequitable access to education, hiring discrimination, and historical redlining practices). Controlling for these characteristics would conceal potential consequences of racism that are important to capture. Therefore, we present unadjusted models, with sensitivity analyses focusing on narrower subpopulations (e.g., full-time workers with at least 6 months job tenure; private sector workers) to interrogate observed disparities.
All analyses were conducted in Stata version 14.2 (College Station, TX: StataCorp LLC). Study procedures were approved by the California Health and Human Services Agency's Committee for the Protection of Human Subjects. Table 1 reports demographic and job characteristics of our sample of Bay Area mothers, grouped by race/ ethnicity. The racial/ethnic composition of our sample is representative of mothers of young children in the Bay Area, with a distribution similar to that in the American Community Survey (ACS) during our study years (5.9% black, 18.5% Hispanic, 31.2% Asian, and 44.4% white in our sample compared to 4.5% black, 12.8% Hispanic, 35.9% Asian, and 44.2% white in ACS). For each characteristic, we report p-values from chi-squared tests.

Results
We observe that most demographic characteristics and all job characteristics vary by race. Black mothers are least likely to be married or cohabiting. Large majorities of white and Asian mothers have a college degree (90.4% of whites and 83.7% of Asians), whereas 33% of black mothers and 36.5% of Hispanic mothers have graduated college. Asian and Hispanic mothers are both much more likely than their counterparts to report speaking a language other than English in the home (37.5% and 41.4%, respectively), and to be born outside of the U.S. Hispanic and black mothers are significantly more likely to be insured by Medi-Cal (California's Medicaid program; 49.5% and 45.4%, respectively) relative to white (4.0%) and Asian (12.7%) women. The number of household children younger than five years does not appear to be associated with race or ethnicity in our sample.
Job characteristics also show strong associations with race/ethnicity among Bay Area mothers. Modest, but significant differences exist in job tenure, with white and Asian mothers being more likely to have worked in their job for at least 5 years and less likely to have a job tenure less than 6 months than black and Hispanic mothers. Similarly, black and Hispanic mothers are less likely to work full time, with only 51.7% of black and 58.3% of Hispanic mothers working over 35 h per week (compared to 82.1% of white mothers, p < 0.001).
We observed a similar pattern for size of employer, with a majority of white and Asian mothers working for large ( > 500 employees) firms, compared to only 28.1% and 34.8% of black and Hispanic women, respectively. White, Asian, and black women work in jobs that are distributed roughly equally between San Francisco and the rest of the Bay Area, but only 31.6% of Hispanic women work in San Francisco. While racial/ethnic differences in job type were not statistically significant, there is a trend toward black women being overrepresented in public sector jobs and underrepresented in private for-profit jobs. We comment on the distribution of our dependent variables by race below. Given these differences in job characteristics, black and Hispanic women may be less likely to be eligible for paid leave programs. In fact, when we examined whether and how much pay women received while on leave, strong racial/ethnic disparities emerged (Fig. 1). Relatively few white (10%) and Asian (13%) women received no pay while on leave, whereas significantly more black (33%; p < 0.01) and Hispanic women (25%; p < 0.05) received no pay. By contrast, only 10% of black women received full pay while on leave, relative to about one-third of white women ( p < 0.05). Despite these differences in access to paid leave, the duration of leave taken was similar across all racial/ ethnic groups (Fig. 1). The majority of women returned to work when their new baby was at least 12 weeks old; very few returned before 6 weeks. We similarly found no significant difference in the likelihood of taking more than 16 weeks (Supplementary Fig. S1).
Lower pay among black and Hispanic women appears to be due to lower likelihood of receiving pay both from the government (e.g., CA-PFL or State Disability Insurance programs) and from employers (Fig. 2). While 71% of white women received 12 to 14 weeks (depending on their mode of delivery) of pay from the government, just about half of black (49%; p < 0.05) and Hispanic (52%; p < 0.01) women received that much pay. In contrast, 47% of black and 36% of Hispanic women did not receive any pay from the government, compared to just 19% of white women ( p < 0.01).
Similarly, while a majority of white women received at least some pay from their employers, 68% of black ( p < 0.01) and 60% of Hispanic ( p < 0.05) women did not receive any pay from their employers. Both groups were significantly less likely to receive at least 12 weeks of paid leave from their employers compared to white women.
Among Hispanic women, these disparities persisted even when restricting our analysis to women employed full time ( > 35 h per week) and who had been employed for at least 6 months (although differences were no longer significance at 5% level due to smaller sample size) ( Supplementary Fig. S2). Black women who had been employed full time for at least 6 months were also significantly less likely to receive any pay from their employer relative to their white peers; however, this subgroup of workers was not significantly less likely to receive government pay during their maternity leave relative to white women.
Employment sector did not explain these differences: disparities in access to employer-paid leave remained for both black and Hispanic women when restricting to those employed by private employers (Supplementary Fig. S3) and private for-profit employers (Supplementary Fig. S4).
Bringing these sources of pay together, we find that almost half of white women (and 43% of Asian women) received both government and employer pay during their leave, significantly more than either black (23%; p < 0.01) or Hispanic women (29%; p < 0.01) (Fig. 3). In contrast, black and Hispanic women were significantly more likely than white women to report not receiving either type of pay (36% and 25%, respectively). Interestingly, 35-36% of women reported government pay alone across all four racial/ethnic groups.
To help understand why we observed racial/ethnic disparities in reported access to paid leave benefits, we examined mothers' knowledge of their maternity leave benefits and the main sources of information about these benefits (Fig. 4). While about threequarters of white and Asian women reported that they understood the maternity leave benefits available to them, significantly fewer black (54%; p < 0.05) and Hispanic (51%; p < 0.01) women did. This likely reflects differential access to information. Black and Hispanic women were significantly less likely than white women to report that their employer helped them learn about their maternity leave benefits (À33 and À31 percentage points, respectively; p < 0.01). In fact, just 63% of black women and 66% of Hispanic women reported that they received help from anyone, compared to 91% of white women ( p < 0.01).
Putting wage replacement level together with the duration of paid leave offered, we developed a measure of FPE weeks of paid leave. White women, on average, received 7.6 FPE weeks of paid leave (Fig. 5). Asian women received slightly less (6.7 FPE weeks). Again, black and Hispanic women received significantly less paid leave: 4.0 and 5.6 FPE weeks, respectively.

Discussion
Consistent with prior research, we find that the duration of leave taken after the birth of a child was similar across racial/ethnic groups; however, the economic implications of that leave differ significantly. We found that black women on average receive 3.6 fewer FPE weeks of paid leave than white women. To quantify that, consider a woman working full time at a minimum wage of $15: in that case, 3.6 fewer weeks of pay amounts to $2160 less over the course of her leave-a large sum for these women.
To put this in perspective, a substantial literature links increased income such as from the earned income tax credit (EITC) with improvements in maternal and infant health. 49 One study finds that an additional $500 per year in EITC reduces the number of bad mental health days by 19% among mothers 50 ; another finds that an additional $500 from a similar Canadian program decreased depression scores by 11% among mothers with a high school degree or less. 51 This magnitude of extra EITC income has also been found to improve birth outcomes, 52 as well as child development. 53 The accumulated evidence suggests that disparities in access to paid parental leave are a preventable source of racial inequities in health that begin at birth.
The reasons underlying disparities in paid leave access are not entirely clear. One hypothesis is that black and Hispanic women are less likely to hold jobs with eligibility for employer-and government-paid leave due to occupational segregation and policy choices that have been made to benefit professional, salaried workers. Part-time workers, for example, are less likely to be eligible for paid leave benefits, 39 and in our sample, black and Hispanic women are overrepresented in part-time work. Unlike many employer policies, CA-PFL program has generous eligibility in terms of hours and job tenure requirements, but excludes most self-employed and public sector workers, again, jobs in which Hispanic and black women, respectively, are overrepresented in our sample.
Another hypothesis is that black and Hispanic women are not taking advantage of benefits that are available to them, potentially because of limited awareness. When we restrict to just private sector workers who are most likely to be eligible for CA-PFL, we still find that 39% of black and 38% of Hispanic women report that they received no paid leave through the government, compared to just 14% of white women.
Our finding that black and Hispanic women were significantly less likely to have received help navigating their benefits from anyone, including, most notably, their employers, suggests that women of color may be missing out on benefits that should be available to them because of inadequate information sharing. The extent to which this differential access to information is rooted in occupational segregation, interpersonal racism and discrimination, or some combination of these factors merits further investigation.
Relatedly, it is possible that black and Hispanic women are more fearful of losing their job if they take leave; although we do not have a clear measure of the importance of this concern, prior analysis has shown that indeed fewer black and Hispanic women are in job-protected positions. 54 Again, the root cause of this differential actual and perceived job security are worthy subjects of future research.
Finally, Hispanic women, many of whom are Spanish speakers, may face language or cultural barriers to accessing information about leave benefits.
The fact that paid leave access depends on the characteristics of one's job suggests that relying on employers to provide this important benefit voluntarily is insufficient for ensuring equity. Government PFML programs are often also similarly deficient, frequently excluding the most economically vulnerable workers, in another manifestation of structural racism.
For example, San Francisco's PPLO provides fully paid leave to eligible workers, but includes only workers employed by private sector employers with at least 20 employees. These restrictions disproportionately exclude low-income workers and completely exclude selfemployed and public sector workers. 48 A more equitable policy approach would resemble Oregon's recently enacted PFML policy, which includes all workers, regardless of firm size or sector, who have earned at least $1000 in wages during a base year and paid into the state's PFML Insurance Fund. 38 Our study addresses a critical gap in the literature by differentiating the source of paid leave and quantifying the proportion of usual pay received while on leave, thus providing a more complete picture of the extent of inequity in paid leave access. Future studies should attempt to replicate and build on these findings in other settings. We focus exclusively on the San Francisco Bay Area, which is demographically, economically, and politically distinct from other regions. Our relatively small sample of black women, although reflective of San Francisco's demography, limits our power to detect small differences.
Future research should also attempt to further disentangle the mechanisms underlying racial/ethnic disparities in paid leave access, including occupational segregation, policy coverage, fear, awareness, and cultural barriers (particularly for Spanish-speaking women).

Disclaimer
The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health or the Robert Wood Johnson Foundation.  Figure S4