Confidential Screening for Sex Trafficking Among Minors in a Pediatric Emergency Department

This study was conducted in an urban, regional, freestanding children’s hospital that serves as a tertiary care level 1 American College of Surgeons–verified trauma center, with a PED census of ∼80 000 patients per year. Licensed psychiatric clinicians, including psychiatrists, are on-site daily, in addition to 24-hour sexual assault nurse examiners, clinical social workers (CSWs), and 17-hours-per-day professional research assistants (PRAs). This study was reviewed and approved by the local institutional review board.

A convenience sample of 12- to 17-year-old English- and Spanish-speaking subjects were enrolled in a prospective cohort between February 2019 and March 2020. Patients were eligible for inclusion if they presented alone, for psychiatric or genitourinary complaints, sexual abuse, sexual assault, or law enforcement concern for trafficking. Patients were excluded if they had altered mental status, hallucinations, developmental or cognitive delays, or minor trauma. PRAs were trained on all study procedures, screened patients, and confirmed eligibility with the treating practitioner. The PRAs approached eligible subjects for enrollment and obtained consent. Accompanying persons were asked to leave the room while the patient filled out the survey to ensure confidentiality and avoid potential traffickers from influencing subject responses. For potential subjects not enrolled because accompanying persons refused to leave the patient room, any clinical concerns for child sex trafficking were addressed through standard evaluation and reporting (CSW consultation, law enforcement [local police department or Federal Bureau of Investigation (FBI)] notification, or Department of Human Services [DHS]). There were no incentives for participation.

Study subjects completed a 6-item confidential screening tool on an electronic tablet in Research Electronic Data Capture.¹⁹  The screening tool includes questions related to physical violence, running away, substance use, sexual history, and previous police involvement (see Supplemental Information). Each question was answered either “yes” or “no,” with the exception of the number of sexual partners, which was categorical (0, 1–5, 6–10, or >10). An answer indicating >5 sexual partners is considered a “yes” response. Subjects who answered “yes” to ≥2 questions on the overall tool were categorized as having a positive screen result.¹⁶  Subjects with a positive screen result completed 4 additional questions exploring behaviors more specific to sex trafficking.

Once the study subject completed the 6-question screening tool, the treating practitioner was informed of the results of the screen (positive or negative) and which questions had positive responses. For subjects who screened positive, the treating practitioner was instructed to consult a CSW or notify the psychiatric CSW to facilitate a psychosocial evaluation for child sex trafficking and report to the appropriate authorities (Fig 1). Treating practitioners completed a separate checklist to collect data on additional risk factors for trafficking (see Supplemental Information); these questions were used in the initial psychometric testing and validation by Greenbaum et al.¹⁶  Once a patient completed the 6-question screening tool, standard care was resumed. The CSW referral to DHS served as the gold standard, establishing the patient as a case of child sex trafficking, although cases were not managed longitudinally through DHS to confirm. If the patient was low risk (<2 questions answered “yes”), then no automatic CSW evaluation was initiated. However, the treating practitioner could request a CSW evaluation at their discretion. For all patients who completed the screening, the treating practitioner indicated the likelihood of trafficking on a Likert scale in the practitioner checklist on the basis of the findings of the CSW evaluation, similar to previous studies^16–18  (Fig 1). Patients were deemed to be positive for risk of trafficking if the CSW determined they were “highly likely” or “likely” to be trafficked. The treating practitioner gave their data collection form directly to a PRA or placed it in a locked box. The PRAs transcribed the information into an online data collection instrument through Research Electronic Data Capture. PRAs notified the primary investigator for all positive screen results to ensure that patient safety assessments were completed by a CSW. If patients had a positive screen result on the tool but denied trafficking behaviors in person, reports were still made if there was reasonable clinical suspicion of trafficking.

For final determination of a study subject’s status, the study primary investigator conducted a retrospective chart review on all true-positives (“highly likely” or “likely” with a positive screen result), false-negatives (“highly likely” or “likely” with a negative screen result), and subjects classified as “unknown.” Two other investigators performed a secondary independent review to reach a final determination.

The data were summarized by using standard descriptive statistics: frequencies and proportions for categorical variables and median and interquartile range (IQR) for non-normally distributed continuous variables. Demographics, screening tool responses, and additional risk factors obtained from the practitioner checklist were compared between patients with a positive and negative trafficking status by using Pearson’s χ² test, Fisher’s exact test, and the Wilcoxon-Mann–Whitney U test. The screening tool’s sensitivity, specificity, positive predictive values (PPVs), and negative predictive values (NPVs) were calculated. Multivariable logistic regression was performed to identify risk factors, in addition to positive screening results on the 6-question screening tool, that were independently associated with child sex trafficking. Significant covariates (P < .05) from the practitioner data collection form were entered into the multivariable logistic regression and remained in the model (P < .1) by using backward selection. All analyses were performed by using SAS version 9.4 (SAS Institute, Inc, Cary, NC).

Power analyses indicated that 575 subjects would need to be enrolled by using an α value of .05 for a trafficking prevalence of 4% or 422 subjects for a prevalence of 11%.

Over the enrollment period, PRAs screened 377 subjects, of which 301 patients were eligible and 214 subjects were enrolled (Fig 2). Two patients had missing data in their practitioner checklist answers regarding the trafficking analysis by the CSW, so 212 subjects were ultimately analyzed. A total of 72.6% of subjects were female, with a median age of 15 years (IQR 13–16; Table 1). Only 4 subjects filled out the survey in Spanish. The most frequent presenting chief complaints were psychiatric (Fig 3).

Out of the 212 patients analyzed, 26 patients were found to be sex trafficked (prevalence 12.3%). A total of 22 of those patients screened positive for trafficking (84.6%), whereas 4 had a negative screen result but were deemed to be trafficked by a CSW (15.4%). A total of 99 patients had true-negative screen results (46.7%), whereas 87 had false-positive screen results (41.0%). All 4 subjects with false-negative results presented with high-risk conditions independently warranting CSW evaluation, whereas most subjects with false-positive results presented with psychiatric complaints or sexual assault, abuse, or violence. The 4 patients with false-negative results are described in Table 2.

The sensitivity and specificity of answering ≥2 questions “yes” on the electronic screening tool were 84.6% and 53.2%, respectively, whereas the PPV was 20.2% (95% confidence interval [CI] 12.7%–27.7%) and NPV 96.1% (95% CI 92.4%–99.9%). An analysis was also completed regarding the sensitivity and specificity of the tool on the basis of the number of questions answered positively (Table 3). ROC (Receiver operating characteristic) curves for the number of questions answered positively on the screening tool are demonstrated in Fig 4. Although the area under the curve was higher for answering 3 questions “yes,” the sensitivity decreased from 84.6% to 80.8%. Positive screening tool responses are shown in Table 4.

Table 5 reveals the additional features of patients who were trafficked versus not trafficked on the basis of the practitioner checklist. Trafficked persons were more likely to be found in a location known for sex trafficking (18.2% vs 0%; P = .01) and more likely to undergo testing for pregnancy, substance abuse, and/or sexually transmitted infections (STIs) compared with nontrafficked subjects. Univariate analyses of the practitioner checklist revealed that sex trafficked minors were more likely to have a history of a previous suicide attempt (38.5% vs 14.0%; P = .004) and have a previous history of abuse or neglect (15.4% vs 2.2%; P = .01) compared with nontrafficked individuals (Table 5). When adding these factors and a positive screen result into a multivariable logistic regression model, only a history of previous suicide attempt was statistically significant (adjusted odds ratio of 3.1; 95% CI 1.2–8.2). However, adding suicide attempt to the screening tool did not significantly alter the test characteristics; after model fitting, the sensitivity remained 84.6% (95% CI 70.8%–98.5%), specificity was 46.8% (95% CI 39.6%–53.9%), PPV was 18.2% (95% CI 11.3%–25.1%), and NPV was 95.6% (95% CI 91.4%–99.8%).

Accurate and timely identification of child sex trafficking in clinical settings has many barriers.²  Standardized screening tools provide an opportunity to identify those at risk for child sex trafficking, although, to date, they have only employed in-person interviewing.^16–18  In this study, we examined the test characteristics of a previously validated 6-question sex-trafficking screening tool administered confidentially via an electronic tablet in English and Spanish. Ultimately, with this method, we accurately identified minors at high risk for child sex trafficking who are in need of additional evaluation and services, congruent with the hypothesis of this study. No additional local risk factors significantly improved the sensitivity of the screen when retroactively fit in a logistic regression model.

With this screening tool administered in a confidential manner, we detected a prevalence of child sex trafficking among high-risk minors of 12.3%, higher than found in previous studies by using in-person interviewing.^16–18  The higher prevalence may be a result of improved reporting by subjects because the disclosure was via self-reporting through an electronic survey. Adolescents and parents agree that screening for risky behaviors during health care visits is important and well accepted,^20–23  including in the PED, whereas electronic formatting of many screening tools integrates a comfortable modality for adolescents to admit risky behaviors.^20,24–29  Winning over the trust of young adolescents is often difficult and may become nearly impossible when a trafficker presents with them to medical care; this confidential screening modality may, therefore, make it easier for minors to answer sensitive questions if their trafficker is present.

Our high prevalence of cases may also be related to the geographic region where the study was conducted: an urban environment with multiple transit points, including a large international airport, quad-directional interstates, cross-country railway systems, large conference centers, and multiple major sports teams. In other studies, researchers examining the geographic distribution of child sex trafficking have found clusters in areas of tourism and transportation hubs as well as in large metropolitan areas where there are high numbers of homeless youth.^13,30,31 

Despite a higher prevalence of child sex trafficking in this study, the sensitivity and NPV of this tool are similar to previous reports,^16–18  suggesting that this is an effective way to screen for child sex trafficking in busy, clinical environments. The survey cutoff points in this study also reflect those in the validation performed by Greenbaum et al,¹⁶  with a similar change in sensitivity and specificity based on the number of questions answered positively. The cases with false-negative results in this population were low. However, those that did have a false-negative screen result still had appropriate evaluations prompted by high-risk behaviors, and, thus, trafficking was still identified.

This screening tool favors sensitivity at the expense of specificity, resulting in the identification of a high number of false-positive results, consistent with other studies.^16,17  This trade-off is acceptable, considering that a failure to identify trafficked persons could result in continued abuse, trafficking, drug abuse, STIs, and possibly even death.⁵  Subjects with false-positive results, ultimately, underwent more testing, however, which is likely warranted, given the high-risk behaviors the subjects endorsed on their positive screen results. Documentation in the medical record should specifically reflect the endorsed high-risk behaviors instead of sex trafficking to decrease any associated stigma. However, because sex trafficking of children <18 years old is child abuse, it is still important to document the full psychosocial evaluation to ensure that other practitioners can be aware of potential trafficking and the minor’s high risk for abuse.

In this population, history of suicide attempt and child abuse increased the odds of screening positive for trafficking. The test characteristics of this screen did not improve when added to a logistic regression model. Similar to other studies, many of the patients presenting with previous trafficking concerns, runaway status, and out-of-control behaviors are triaged as primary psychiatric complaints, resulting in a high proportion of these patients in this study population.^{1,6,12,15,18,32}  Therefore, a history of suicide attempt may not have improved the screening tool because of the high proportion of psychiatric complaint enrollment because patients may have been suicidal for reasons not related to trafficking. Similarly, a history of child abuse can be associated with running away from home, a screening question on the tool, which may have increased the odds of trafficking seen here also.

Over the course of this study, some patients contradicted their screening responses when practitioners or CSW asked additional questions in person. Although the confidential tablet survey was one of the strengths of this screening method,^20,24  it became unclear how to report findings to DHS when the patient verbally denied their behaviors. Subjects, often, do not identify themselves as victims, a well-described phenomenon of exploited persons described in other studies³²  and, therefore, may be less likely to admit to victimization in person. Therefore, in this study, investigators adapted the research procedures to use the tablet survey responses as the patient’s admission of their behaviors. Although this was a small number of patients, these cases were reported to DHS.

The true prevalence of trafficked persons is largely unknown.⁵  Generally, it is difficult to obtain the true prevalence and incidence of trafficking because of the secretive nature of the process: trafficked individuals are trained to not disclose information to protect their traffickers from arrest and litigation, and trafficked persons are led to believe that they may also be arrested or prosecuted for their behaviors. “Trauma bonding” may also facilitate the protection of the trafficker by the trafficked person because the trafficked person sees the attentions paid to them by their trafficker as affection, despite actual physical, psychological, and sexual abuse.³³  Finally, because suicide screening in PEDs is becoming more prevalent in the general adolescent population,^20,23,34  which is well tolerated by patients and practitioners while improving suicide detection, it would be prudent to also begin screening the general adolescent population for trafficking to determine the true prevalence.

This study has certain limitations. Overall, only 4 Spanish-speaking patients were enrolled, although a large percentage of the local population is Spanish-speaking only. Although this was not the object of study, we acknowledge that there may be unknown cultural factors that may decrease detection of these high-risk individuals. In addition, bilingual subjects may have selected English surveys despite the available Spanish version. Further study is required to explore additional cultural barriers.

Given that CSW assessments are often lengthy, a final determination of the patient’s trafficking status was not always available. This may be, in part, due to the transfer of psychiatric patients to other inpatient hospitals; therefore, follow-up CSW notes were not available for investigator review. Thus, using the referral to DHS via CSW as the gold standard, in this study, we may have overestimated the prevalence of trafficking in this population. However, the prevalence, although higher, is within the range of that found in other studies conducted in large urban environments and, therefore, reflective of the population.¹⁷ 

Finally, the severe acute respiratory syndrome coronavirus 2 pandemic interrupted all study protocols at the study site, limiting enrollment. However, the study sample size is similar to previous reports^16–18  and achieved comparable test characteristics, suggesting that the sample enrolled adequately represents the local high-risk population.

Screening for child sex trafficking in high-risk minors on a confidential tablet is an accurate and sensitive method of identifying patients at high risk for child sex trafficking in a busy PED. This modality of screening may be helpful in other clinical environments to decrease the burden of requiring in-person, trauma-informed interviews to screen for trafficking and improve appropriate care for this population.

We acknowledge those who have contributed their time and energy toward the completion and success of this study: the PRAs in our emergency department for enrolling patients; Alisa Thomas, Megan Cook, and Lisa Justis from our clinical social work department for their assistance in coordinating CSWs to our study; Dr Beau Carubia and Dru Hunter from the department of psychiatry in the emergency department for assisting the clinical psychiatrists and social workers’ evaluations, education, and response; and Dr Jordan Greenbaum for her continued support and guidance in using her screening tool.

CI

confidence interval

CSW

clinical social worker

DHS

Department of Human Services

FBI

Federal Bureau of Investigation

IQR

interquartile range

NPV

negative predictive value

PED

pediatric emergency department

PPV

positive predictive value

PRA

professional research assistant

STI

sexually transmitted infection