Reducing the time from human immunodeficiency virus (HIV) infection to diagnosis is central to efforts to control the HIV epidemic. Early diagnosis and treatment initiation are associated with improved clinical outcomes and reduced risk of transmission to susceptible partners.1,2 In 2015, an estimated 15% of persons living with HIV in the United States were undiagnosed, and the median time from infection to diagnosis was 3 years.3 To improve primary and secondary prevention endpoints, more work is needed to define and improve implementation of effective strategies to identify individuals earlier in the course of infection.
The Centers for Disease Control and Prevention (CDC) and the US Preventative Services Task Force recommend that all adults and adolescents undergo screening for HIV at least once in their lifetime regardless of risk.4,5 This recommendation for opt-out routine screening is in addition to guidelines calling for screening of high-risk persons at least annually, screening the sex and injection partners of diagnosed cases, screening pregnant women as part of prenatal care, and testing patients with HIV/AIDS-associated symptoms. Modeling studies have suggested that 1-time routine screening will be cost-effective in settings where the prevalence of undiagnosed infection exceeds 0.05% to 0.2%,6 and national guidelines recommend routine screening unless the undiagnosed prevalence is less than 0.1%.4
However, implementation of routine screening has been suboptimal,7 and evidence of the impact on case-finding and prevention outcomes has been mixed. In the years following the 2006 revision to the CDC HIV testing guidelines, the proportion of people who had ever tested for HIV in the United States remained largely unchanged.8 Although some evaluations of routine screening programs have reported increases in diagnoses,9,10 others have reported equivalent or lower case detection compared with targeted and symptom-driven testing strategies.11–13 To guide effective implementation of routine screening and identify settings where it is likely to be impactful, analyses need to account for existing patterns of screening and the risk profile of the population.
In this article, we present the findings from a mathematical model developed to evaluate the population-level impact of adding routine screening to other established screening strategies in terms of case detection, cumulative years of undiagnosed infection, and the number of cases that develop HIV/AIDS-associated symptoms before diagnosis. To highlight the influence of local epidemic context on the impact of routine screening, we fit 3 models: a national model, a model representing a jurisdiction with infection concentrated among men who have sex with men (MSM) (King County, WA), and a model representing a jurisdiction with a more diverse distribution of infection (Philadelphia County, PA). To inform more specific and impactful guidelines, we estimated the ages for 1-time routine screening at which diagnostic yield is maximized in each setting. We compared the incremental impact of routine screening at these optimal ages to the impact of risk-based screening on prevention and clinical outcomes.
We constructed a static linear mathematical model using Microsoft Excel (Version 16.9) to represent the processes of HIV infection, testing, and diagnosis in a hypothetical population of 100,000 persons. To account for heterogeneity in HIV risk and patterns of testing, we divided the population into 4 groups: high- and low-risk MSM, men who have sex with women only (MSW), and women. All persons enter the model at age 16 years and progress in yearly time steps through age 64 years or HIV diagnosis, at which point they exit the model.
Our model evaluates the incremental impact of adding 1-time routine screening to 4 baseline targeted HIV testing strategies: (1) repeat screening of those at high risk of infection (risk-based screening); (2) prenatal screening; (3) testing prompted by partner notification; and (4) diagnostic testing at the onset of HIV/AIDS-associated symptoms. As described below, the proportion of persons with indications for targeted testing who are tested via these strategies was informed by data on symptom onset and screening practice. Because repeat HIV testing among MSW and women is not common,14 our model represents risk-based screening only among MSM. Routine screening is implemented at specific ages and offered to all persons who have not engaged in risk-based or prenatal screening in that year.
To account for uncertainty in parameter estimates, we conducted Monte Carlo simulation using @RISK software (version 7.6; Palisade Company, Ithaca, NY). Distributions were defined based on the range of estimates in the literature or expert opinion (see Table 1 and Supplemental Digital Content 1, http://links.lww.com/OLQ/A471, which includes additional details on model parameterization). Additional details about the model, including a model schematic and equations describing the modeled processes and outcomes, are presented in the supplemental appendix (Supplemental Digital Content 1, http://links.lww.com/OLQ/A471). The model was parameterized using summary statistics from previous analyses and does not qualify as human subjects research.
Parameters and Data Sources
Model inputs and sources are summarized in Table 1, and references are listed in Supplemental References (Supplemental Digital Content 2, http://links.lww.com/OLQ/A472). To parameterize the 2 subnational models, we used data from King and Philadelphia Counties to define the risk group distribution, HIV incidence rates, and birth rates. All other parameters were held at the values used in the national model. We refer to the model informed by King County data as the concentrated epidemic (CE) model, and the model informed by Philadelphia data as the diverse epidemic (DE) model.
Risk Group Distribution
To distribute the modeled cohort by gender, we used American Community Survey (ACS) data on the number of males and females ages 16 to 64 years at the national level and in King and Philadelphia Counties.31s,32s For each setting, we applied estimates of the proportion of MSM in the past 5 years to define the size of the MSM population.33s,34s Uncertainty intervals (UIs) for this proportion were defined to encompass 95% confidence intervals and estimates from other sources.35s The percentage of MSM who are high risk was informed by estimates of the percent with indications for use of HIV preexposure prophylaxis.36s
HIV Incidence Rates
For the national model, we used published estimates of the number of incident HIV cases in 2016 attributable to heterosexual contact among men and women and to male-male sex, including MSM who also reported injection drug use.37s The age distributions of infection for MSM, MSW, and women were defined as the weighted average of race/ethnicity-specific distributions. We derived the age distribution specific to MSW cases from the calculated distributions for all males and MSM. We used the 2016 ACS data31s to define denominators for calculation of age-specific incidence rates, assuming the proportion of males who are MSM to be uniform across age groups.
For the CE model, we obtained estimates of the number of incident infections in King County in 2016 from a back-calculation method developed to measure undiagnosed infection38s (J.K. Birnbaum, personal communication). Incident infections were estimated for MSM and other noninjection adult cases, and we apportioned the latter by gender according to the distribution of new diagnoses.35s For the DE model, we used published Philadelphia County incidence estimates for 2016.39s Lacking data on the age distribution of infection in these jurisdictions, we applied the same age distributions as in the national model. We defined population denominators using the 2016 ACS data in combination with estimates of MSM population sizes, as described above.
In all 3 models, we assumed that all infections in individuals 13 years or older would occur between the ages of 16 and 64 years. Incidence rates for MSM were calculated, assuming the incidence rate among high-risk MSM to be 2.5 times that of low-risk MSM,40s with a uniform UI of 2 to 3 times higher. With these inputs for incidence, the mean percentage of cases that are MSM was 71% in the national model, 77% in the CE model, and 57% in the DE model. The lifetime risk of infection was 1:135, 1:225, and 1:62 for the 3 models, respectively.
We modeled patterns of risk-based screening by dividing the MSM population into 3 groups: men who do not engage in risk-based screening, men who screen every year (as recommended), and men who screen less than once a year. The screening interval for nonannual screeners was sampled from a normal distribution with a mean of 3 years (Table 1). The proportion of men in each group varied by age, and high-risk MSM were more likely to engage in risk-based screening than low-risk MSM.41s To examine the sensitivity of our results to different levels of risk-based screening, we defined scenarios with low, mid-range, and high engagement in risk-based screening. For the mid-range scenario, the proportion of men in each risk-based testing group was defined based on the data from national surveys of MSM.42s-45s In the low risk-based screening scenario, the proportion of men who screen annually was reduced, and the proportion who do not regularly screen was increased, both by 30%. In the high risk-based screening scenario, annual screening was increased, and the no-screening group was reduced by 30%. Figure 1 displays the overall proportion of men in each screening group by age for the low, mid, and high screening scenarios.
Birth rates were used to define the number of women who become pregnant at each age.46s-48s We used published data to define the proportion of pregnant women screened for HIV.49s-52s
Partner Notification Testing
We sampled values for the number of additional persons diagnosed through partner notification per index case from a normal distribution based on the range of published estimates.53s-56s
Following the approach adopted by Golden et al,15 we assumed that 8.8% of persons develop a CD4+ cell count <200 cells/mm3 within the first year of infection, after which the risk of dropping below this threshold is linear at an annual rate of 6.08%.57s Symptom onset was assumed to coincide with CD4+ depletion, prompting diagnostic testing.
Routine Screening Coverage
For consistency with previous models of routine screening,16,17 we assumed that 80% of eligible individuals are offered and accept routine screening.
To identify the optimal age for routine screening, we examined the impact of adding 1-time screening at each age in the model. For each specified age, all persons not tested through risk-based or prenatal screening in that year were considered eligible for routine screening. Outcomes were defined as the percentage of tests that result in a new diagnosis (test positivity), the population-level person-years of undiagnosed infection, and the number of cases that progress to symptomatic HIV/AIDS before diagnosis. We assessed that the optimal ages for routine screening in each setting as the range of ages for which test positivity was within 10% of the maximum value.
For comparison with targeted screening strategies, we evaluated the incremental impact of implementing routine and risk-based screening. In a baseline scenario, diagnosis occurred only through symptom-based testing, prenatal screening, or partner notification. We then measured the impact of adding low- and high-level risk-based screening to these strategies. The third and fourth scenarios added routine screening at the optimal ages to scenarios with low- and high-level risk-based screening, respectively. For each scenario, we report the test positivity for the added screening strategy and the incremental percent change in person-years of undiagnosed infection and symptomatic HIV/AIDS cases relative to the baseline scenario.
In all analyses, we assumed 100% HIV test sensitivity and no window period. Outcomes are presented as the mean and 95% UIs from 500 Monte Carlo simulations.
Optimal Age for Routine Screening
In the national-level model, the test positivity from routine screening increased with age up to a peak of 0.06% (95% UI: 0.05%, 0.07%) at 31 years in a scenario with mid-range risk-based screening (Fig. 2). Screening between the ages of 26 and 36 years resulted in test positivity within 10% of this peak value (Table 2). The optimal age for routine screening was lower in the CE model, with test positivity maximized at 0.04% (95% UI: 0.04%, 0.05%) at age 26 years (note that y-axes differ across plots in Fig. 2). In the DE model, routine screening at age 44 years yielded a test positivity rate of 0.16% (95% UI: 0.14%, 0.19%). The age ranges for which test positivity was within 10% of the optimal value were 24 to 33 years and 38 to 52 years in the CE and DE models, respectively.
By gender, test positivity was projected to be higher and maximized at younger ages for men than women across modeled settings (Fig. 2). Routine screening yielded the most diagnoses at age 44 years for women in all settings, but the optimal ages for men depended on the proportion of cases in MSM. In the national and CE models, where MSM accounted for greater than 70% of prevalent cases, peak routine screening test positivity for males occurred at ages 28 and 25 years, respectively, in the context of mid-range risk-based screening among MSM. In the DE model, where MSM accounted for an estimated 57% of prevalent cases, peak test positivity for men was at age 44 years.
Relative Impact of Routine and Targeted Screening Strategies
Risk-based screening test positivity ranged from 0.22% (95% UI: 0.18, 0.28) with high-level screening in the CE model to 1.20% (95% UI: 0.87%, 1.58%) with low-level screening in the DE model (Table 3). Relative to a scenario with symptom-based, prenatal, and partner notification testing, the incremental impact of adding low-level risk-based screening in the national model was a 51% (95% UI: 46%, 56%) reduction in person-years of undiagnosed infection and a 57% (95% UI: 52%, 63%) reduction in symptomatic HIV/AIDS cases. With high-level risk-based screening, these outcomes were reduced by 62% (95% UI: 59%, 65%) and 69% (95% UI: 64%, 72%), respectively. In the CE model, the incremental impact of risk-based screening was of a similar magnitude. The impact was slightly lower in the DE model, with mean estimates of 41% to 50% reductions in person-years of undiagnosed infection and 46% to 55% reductions in symptomatic cases (Table 3).
Test positivity with routine screening in the national model was 0.07% (95% UI: 0.06%, 0.09%) when added to low-level risk-based screening and 0.05% (95% UI: 0.04%, 0.06%) when added to high-level risk-based screening (Table 3). The corresponding mean estimates for routine screening test positivity were 0.04% and 0.03% in the CE model, and 0.17% and 0.15% in the DE model. Incorporating UIs, the percent reduction in person-years of undiagnosed infection with routine screening across all 3 models ranged from 3% to 8% and the percent reduction in symptomatic cases ranged from 3% to 11%. For comparison, increasing risk-based screening from low to high levels in the national model resulted in an 11% reduction in person-years of undiagnosed infection (95% UI: 8%, 13%) and an 11% reduction in symptomatic HIV/AIDS cases (95% UI: 8%, 14%). These gains were achieved with an average of 4.75 times fewer screening tests than in the scenario adding routine screening to low risk-based screening (data not shown).
By accounting for heterogeneity in HIV risk and explicitly representing established targeted testing practices, our model provides context-specific estimates of the potential impact of routine screening in different settings in the United States. Our results suggest that routine screening may be an effective case-finding strategy, particularly in settings with high HIV incidence and a relatively high proportion of cases in women and non-MSM men, such as Philadelphia County. However, even at optimal ages with high test acceptance, our model indicates that 1-time routine HIV screening in the general population will have limited impact on cumulative years of undiagnosed infection and progression to symptomatic HIV/AIDS before diagnosis.
We found that the optimal ages for routine screening depend on the local epidemic context. In the setting with HIV infection concentrated in MSM, the optimal age range was younger than that in the setting with a more diverse epidemic. The range of ages resulting in test positivity within 10% of the optimal value was also narrower with a more concentrated epidemic, driven by heightened incidence at younger ages for MSM. At the national level, our results are comparable to a previous model of the optimal age for routine screening in adults and adolescents15 and reinforce the conclusion that 1-time screening before age 24 years is not the most efficient use of resources.17
Consistent with previous models,16,18 our findings highlight a tradeoff between targeted and nontargeted screening. With higher practice of risk-based screening, the incremental impact of adding routine screening is diminished. However, the relative benefits of risk-based and routine screening depend on the epidemiologic context. In settings with transmission largely concentrated in MSM, increasing from low- to high-level risk-based screening resulted in shorter average time to diagnosis and fewer symptomatic cases than adding routine screening. In the DE model, these scenarios were more comparable, although increasing risk-based screening required far fewer screening tests to achieve the same results as routine screening in all 3 modeled settings. Of note, our outcomes do not account for the effects of screening on HIV transmission resulting from adoption of preventative measures, such as preexposure prophylaxis, changes in risk behavior, or earlier initiation of treatment.1,2 Taking these effects into account would likely amplify the relative population-level benefits of targeted screening strategies, because improving prevention and treatment in groups with high rates of partner turnover and among pregnant women would avert more infections. However, the optimal ages for routine screening might shift, as the ages with the highest prevalence of undiagnosed infection may not correspond to the ages of peak transmission risk.
A key advantage of routine screening is that it does not depend on risk assessment and captures people who are not the typical focus of targeted testing campaigns.11,12 Survey data indicate that as many as 39% of MSM do not disclose to their health care providers that they have sex with men,19 and providers do not always offer HIV tests to those with indications for repeat testing.20 As a result, many high-risk individuals, in addition to those who do not engage in high-risk behaviors, fall through the cracks with targeted screening.21 If routine screening detects cases in historically underscreened groups, it may help reduce demographic and geographic disparities in HIV outcomes. However, evaluations of routine screening programs have reported low rates of testing,12,22 with perceptions of low risk reported as a common reason for screening not being offered or accepted.9,23 Perceived and actual risk are often misaligned,24 such that overreliance on opt-out routine screening may miss opportunities to identify cases25 and engage persons at risk of infection in prevention strategies like preexposure prophylaxis. More work is needed to improve clinic procedures to prompt and facilitate both routine and risk-based screening, as has been successfully implemented in some settings.13
An outcome of interest in evaluating routine screening programs is test positivity. We found that test positivity exceeded the established 0.1% threshold of cost-effectiveness4 only in the DE model. However, our analysis does not account for within-region heterogeneity in patient populations. For example, persons who present at emergency departments or inpatient hospitals may be more likely to have undiagnosed HIV, and some evaluations of routine screening programs in these clinical settings have reported a test positivity greater than 0.1%.9,12,22 Additionally, this threshold for cost-effectiveness may have shifted with adoption of test-and-treat policies,26 and willingness to pay may be higher as jurisdictions work toward meeting the goals of the Ending the HIV Epidemic Initiative. Furthermore, our model does not consider the potential benefits of routine screening in identifying previously diagnosed persons who may be out of care.10
An additional limitation to our model is the dependence on estimates of HIV incidence, which is measured with substantial uncertainty. A recent analysis suggested that HIV cases are overcounted by 26% in New York City and may be overcounted by as much as 100% to 200% in other jurisdictions.27 King County28 and Philadelphia County implement a rigorous deduplication procedure, but incidence estimates at the national level and from other settings may be biased, resulting in an overestimation of test positivity. Improvements in surveillance systems and methods to estimate incidence could improve the accuracy of model predictions. Data on age of infection are also subject to a high degree of uncertainty, and small numbers for some subgroups make these estimates unreliable.37s Lacking data on the risk-group-specific age distribution of infection in King and Philadelphia Counties, we assumed the age distributions to be the same as at the national level. This assumption may not be valid, so future models should seek to incorporate data on how age at infection varies across contexts. Incidence rates are also influenced by uncertainty in the size of the MSM population, although results produced by sampling within a range of plausible values were not meaningfully different.
Our model made a number of simplifying assumptions. Based on data indicating low levels HIV testing in non-MSM populations,14 we modeled risk-based screening only in MSM. To the extent that females and MSW engage in serial risk-based screening, our model will overestimate the impact of routine screening. We assumed that diagnostic testing occurs for all individuals with progression to CD4 count less than 200 cells/mm3. Although we modeled heterogeneity in the time to reach this immunologic threshold, we do not fully capture variability in the timing of symptom onset and resulting diagnosis. However, we do not expect variability in this process to meaningfully affect modeled outcomes. We also did not account for transmission attributable to injection drug use among non-MSM men and women, which contributed 5% of incident cases in 2016.37s Although injection drug users engage in risk-based screening,29 recent increases in diagnoses in this population30 point to the importance of improving case finding and may add value to routine screening programs.
Our findings highlight the importance of developing diverse and balanced HIV testing programs. Even with a relatively modest impact on HIV prevention and care outcomes, making HIV testing a routine component of health care provision is valuable to help normalize HIV testing and to detect cases that are missed by targeted testing strategies. Together with estimates from previous models,15,17 our findings can be used to inform more efficient and impactful implementation of routine screening at optimal ages in different contexts. Ultimately, whether routine screening is indicated in any particular clinical setting is an empirical question, but at the population level, our model suggests that the impact of routine screening will be modest in comparison with more targeted testing strategies. Thus, it will be important to ensure that implementation of routine screening does not detract from targeted testing efforts.
1. Saag MS, Benson CA, Gandhi RT, et al. Antiretroviral drugs for treatment and prevention of HIV infection in adults: 2018 recommendations of the international antiviral society—USA panel. JAMA 2018; 320:379–396.
2. Khosropour CM, Dombrowski JC, Kerani RP, et al. Changes in condomless sex and serosorting among men who have sex with men after HIV diagnosis. J Acquir Immune Defic Syndr 2016; 73:475–481.
3. Dailey AF, Hoots BE, Hall HI, et al. Vital signs: Human immunodeficiency virus testing and diagnosis delays—United States. MMWR Morb Mortal Wkly Rep 2017; 66:1300–1306.
4. Branson BM, Handsfield HH, Lampe MA, et al. Revised recommendations for HIV testing of adults, adolescents, and pregnant women in health-care settings. MMWR Recomm Rep 2006; 55:1–17; quiz CE1–4.
5. Owens DK, Davidson KW, Krist AH, et al. Screening for HIV infection: US preventive services task force recommendation statement. JAMA 2019; 321:2326–2336.
6. Walensky RP, Freedberg KA, Weinstein MC, et al. Cost-effectiveness of HIV testing and treatment in the United States. Clin Infect Dis 2007; 45(Suppl 4):S248–S254.
7. Herrin J, Wesolowski LG, Heffelfinger JD, et al. HIV screening practices and hospital characteristics in the U.S., 2009–2010. Public Health Rep 2013; 128:161–169.
8. Hall HI, Song R, Tang T, et al. HIV trends in the United States: Diagnoses and estimated incidence. JMIR Public Health Surveill 2017; 3:e8.
9. Lyons MS, Lindsell CJ, Ruffner AH, et al. Randomized comparison of universal and targeted HIV screening in the emergency department. J Acquir Immune Defic Syndr 2013; 64:315–323.
10. Rodriguez V, Lester D, Connelly-Flores A, et al. Integrating routine HIV screening in the New York City Community Health Center Collaborative. Public Health Rep 2016; 131(Suppl 1):11–20.
11. Klein PW, Messer LC, Myers ER, et al. Impact of a routine, opt-out HIV testing program on HIV testing and case detection in North Carolina sexually transmitted disease clinics. Sex Transm Dis 2014; 41:395–402.
12. Osorio G, Hoenigl M, Quartarolo J, et al. Evaluation of opt-out inpatient HIV screening at an urban teaching hospital. AIDS Care 2017; 29:1014–1018.
13. Haukoos JS, Hopkins E, Bender B, et al. Comparison of enhanced targeted rapid HIV screening using the Denver HIV risk score to nontargeted rapid HIV screening in the emergency department. Ann Emerg Med 2013; 61:353–361.
14. Pitasi MA, Delaney KP, Oraka E, et al. Interval since last HIV test for men and women with recent risk for HIV infection—United States, 2006–2016. MMWR Morb Mortal Wkly Rep 2018; 67:677–681.
15. Golden MR, Hughes JP, Dombrowski JC. Optimizing the timing of HIV screening as part of routine medical care. AIDS Patient Care STDS 2017; 31:27–32.
16. Paltiel AD, Weinstein MC, Kimmel AD, et al. Expanded screening for HIV in the United States—An analysis of cost-effectiveness. N Engl J Med 2005; 352:586–595.
17. Neilan AM, Dunville R, Ocfemia MCB, et al. The optimal age for screening adolescents and young adults without identified risk factors for HIV. J Adolesc Health 2018; 62:22–28.
18. Shah M, Risher K, Berry SA, et al. The epidemiologic and economic impact of improving HIV testing, linkage, and retention in care in the United States. Clin Infect Dis 2016; 62:220–229.
19. Durso LE, Meyer IH. Patterns and predictors of disclosure of sexual orientation to healthcare providers among lesbians, gay men, and bisexuals. Sex Res Social Policy 2013; 10:35–42.
20. Frasca K, Nassau T, McLaughlin C, et al. Factors associated with recent HIV testing among men who have sex with men in Philadelphia: A cross-sectional analysis of the National Behavioral Surveillance System Survey. AIDS Care 2018; 1–8.
21. Weissman S, Kintziger K, Albrecht H, et al. Late HIV diagnosis and missed opportunities for HIV testing in South Carolina. Presented at: Conference on retroviruses and opportunistic infections . 2016; Boston, Massachusetts.
22. Branson BM, Chavez PR, Hanscom B, et al. Expanding hospital human immunodeficiency virus testing in the Bronx, New York and Washington, District of Columbia: Results from the HPTN 065 study. Clin Infect Dis 2018; 66:1581–1587.
23. Shirreffs A, Lee DP, Henry J, et al. Understanding barriers to routine HIV screening: Knowledge, attitudes, and practices of healthcare providers in King County, Washington. PLoS One 2012; 7:e44417.
24. Blumenthal J, Jain S, Mulvihill E, et al. Perceived versus calculated HIV risk: Implications for pre-exposure prophylaxis uptake in a randomized trial of men who have sex with men. J Acquir Immune Defic Syndr 2019; 80:e23–e29.
25. Czarnogorski M, Brown J, Lee V, et al. The prevalence of undiagnosed HIV infection in those who decline HIV screening in an urban emergency department. AIDS Res Treat 2011; 2011:879065.
26. Lucas A, Armbruster B. The cost-effectiveness of expanded HIV screening in the United States. AIDS 2013; 27:795–801.
27. Xia Q, Braunstein SL, Wiewel EW, et al. Persons living with HIV in the United States: Fewer than we thought. J Acquir Immune Defic Syndr 2016; 72:552–557.
28. Hood JE, Katz DA, Bennett AB, et al. Integrating HIV surveillance and field services: Data quality and care continuum in King County, Washington, 2010–2015. Am J Public Health 2017; 107:1938–1943.
29. Broz D, Wejnert C, Pham HT, et al. HIV infection and risk, prevention, and testing behaviors among injecting drug users—national HIV Behavioral surveillance system, 20 U.S. cities, 2009. MMWR Surveill Summ 2014; 63:1–51.
30. Department of Health and Human Services, Centers for Disease Control and Prevention. NCHHSTP AtlasPlus. 2017. Available at: https://www.cdc.gov/nchhstp/atlas/index.htm
. Accessed 12 March 2019.
For additional references see “Supplemental References,” http://links.lww.com/OLQ/A472.