Comparison of modes of administration of screens to identify a history of childhood physical abuse in an adolescent and young adult population

Background— Childhood physical abuse is a major public health issue with negative consequences to health and well-being manifested in childhood and adolescence, and persisting into adulthood. Yet much childhood physical abuse is not identified when it occurs and little is known about how to screen for it. Methods— To address this gap, the effectiveness of four modes of administration of screens to identify childhood physical abuse were compared in a sample of 506 adolescents and young adults ages 12 – 24 years seeking general health services at a primary care clinic. Comparisons were made between paper and pencil screen, Audio Computer Assisted Self Interview (ACASI) screen, face to face structured screen (all three utilizing the same measure), and face to face unstructured interview. Results— Overall, 44.5% of the sample disclosed that they had been physically abused. When compared to paper and pencil screen, the odds of reporting physical abuse was 1.5 (95%CI: 0.92, 2.58) and 4.3 (95%CI: 2.49, 7.43) higher among participants using face-to-face structured screen and face-to-face unstructured interview methods, respectively. The face-to-face unstructured interview identified significantly higher reports than the paper and pencil screen. Conclusions— While the unstructured interview was the most effective mode for screening for childhood physical abuse, additional research is needed to confirm whether this holds true in other health care settings. Further research should examine how a health provider’s training, experience, and comfort level might influence the identification of physical abuse disclosure in primary care settings using face to face unstructured interview.


Introduction
Childhood physical abuse is a major public health issue with tremendous emotional and financial burden. 1 Though much abuse goes unreported 2 the number of reported cases among children and adolescents nationally is high: in 2013 there were 3.5 million reports of child maltreatment involving 6.4 million children of which 18% were for physical abuse. 3 Childhood physical abuse has both short and long-term negative consequences which affect all aspects of functioning throughout the victim's life course. 2,4,5 In adolescents, the problems associated with abuse include teen pregnancy, 6 high stress, poor self-esteem, cigarette smoking, drug and alcohol abuse, 7,8 and depression and suicidality. 9 These negative effects can be diminished through treatment interventions if the abuse is identified by a health care provider. 1,2,10,11 While most victims do not spontaneously disclose a history of childhood physical abuse, they are likely to disclose if asked in a medical setting as part of a comprehensive health history. 12,13,14 Unfortunately most health care providers do not ask about abuse when there are no obvious signs or symptoms, as is most commonly the case. 15 Though very few studies have focused on understanding why providers do not assess for childhood abuse, 16 there is evidence that they feel ill-prepared and lack the knowledge of effective methods for identification. 18,17 A number of modes of administration of screens have been used to identify a history of childhood abuse including paper and pencil questionnaires, interviewer conducted questionnaires, computer assisted questionnaires, and face to face interviews. 18 Each has its merits. The paper and pencil questionnaire is easy to administer but depends on the reader understanding and correctly interpreting questions. 19 In contrast the Audio Computer Assisted Self Interview (ACASI) has an audio component which speaks the questions to the participant and does not require the same level of reading skills. 22 Structured screens, such as the Childhood Maltreatment Interview Schedule-Short Form (CMIS-SF) 20 or the Computer Assisted Maltreatment Inventory (CAMI) 21 use a defined set of questions. In contrast, the face to face unstructured interview allows the give and take of a conversation 22,23 allowing the interviewer to probe. Thus, an experience of physical punishment that a participant might initially define as non-abusive might upon further probing become re-defined as abuse. ACASI, which has not previously been studied in childhood abuse per se, has been found to be more effective than other modes of inquiry in research on highly sensitive issues in adolescents and young adults, 24,,25,26,27,28,29 as it has also been shown to enhance the participants' sense of privacy and to reduce the influence of social desirability in shaping participants' responses. 30 Our aim was to compare the effectiveness of four modes of administration of screens -paper and pencil screen, ACASI screen, face to face structured screen, and face to face unstructured interview -to identify a history of childhood physical abuse, during a clinical visit.

Study Population
The study sample was recruited from English speaking youth ages 12 to 24 years, seeking general health services, between December 5, 2005 and April 13, 2007, at a New York City primary care clinic specifically designed for young people. A total of 532 young people were screened for history of childhood physical abuse.

Study Recruitment
Institutional Review Board approval was obtained from the Icahn School of Medicine at Mount Sinai along with a waiver of parental consent to allow consent from adolescents under age 18. A certificate of confidentiality was obtained to protect participants' privacy.
While waiting to see their medical provider, patients were approached by a research assistant who described the project as a confidential study on how to best take a psychosocial history from young people. Patients were told that they could decide against participation at any time without this affecting their care. Those with difficulty understanding the study materials and consent form were not enrolled. No formal sampling or selection protocol was used. Patients who agreed to participate, once consented, were randomized within clinician and non-clinician arms to one of four modes of administration of screens to identify a history of childhood physical abuse. Participants received two movie tickets upon completion of all the study instruments. Safety protocols were put in place to ensure an immediate assessment for any participant who disclosed childhood abuse or suicidality. For those under 18 years who disclosed abuse, child protection reporting protocols were followed.

Study Randomization
The study was limited by the fact that only one clinician was assigned to conduct the two face-to-face screening groups. Therefore, random allocation was stratified based on clinician's availability. When the clinician was not available, participants were randomized to paper and pencil screen versus ACASI screen and when the clinician was available participants were randomized to face to face structured screen versus face to face unstructured interview.

Outcome
The study outcome was self-reported history of childhood physical abuse occurring before 17 years of age disclosed during any of the three structured screening methods (paper and pencil, ACASI or face to face structured screens) or a face to face unstructured interview.
The outcome was specified as childhood physical abuse or no childhood physical abuse regardless of the screening method used. For all three structured methods, childhood physical abuse was identified using the Childhood Maltreatment Interview Schedule-Short Form (CMIS-SF) (see Appendix) modified to better fit the speech used by the study population.

Predictors
Once participants completed the history of childhood abuse using one of the four randomly assigned modes of administration of childhood abuse screens, the participants completed a demographic questionnaire and the Beck Depression Inventory for Primary Care-Fast Screen (BDI-FS) 31 using ACASI.
The primary predictor of interest is the mode of screening to identify a history of childhood physical abuse. The following covariates: age, gender, race, ethnicity, zip code, nativity status (immigration status), last grade completed, school enrollment status, school performance, and living arrangement most of the time within the last year) were considered as potential confounders and were adjusted for in the statistical model.

Statistical Analysis
The statistical analysis was conducted by Viswanathan Shankar, DrPH. The distribution of socio-demographic variables was presented as frequencies and percentages and bivariate associations were examined using the Pearson chi-square statistics.
Approximately, five percent of the covariates had missing information, thus we modeled the data both as complete case data (n=506) and as a multiply imputed data (532 × 10 dataset). Multiple imputation was done using fully conditional specification methods which is a flexible imputation procedure that models incomplete variables by a set of conditional densities using different regression procedure. Ten imputation datasets were created with 200 burn in iterations under the missing at random (MAR) assumptions.
Multivariable logistic regression models were fitted to examine the effect among the modes of administration and physical abuse status after adjusting for potential confounders for both complete case and multiple imputation data. Potential covariates that were associated with the outcome at 20% level were selected for final models.
All analyses were performed using SAS software 9.4. 32

Results
The distribution of participant characteristics by study arms (modes of screen) are presented in table 1. Over half of the participants were age 18 and older (52.2%). Most were female (85.3%), Hispanic/Latino or black (93.3%), and almost a third resided in Harlem (32.7%). The majority were U.S. born (81.6%), currently in school (79.5%) and most had graduated from high school or were still in school at the right grade for their age (88.3%). More than a quarter of participants (27.2%) were found to have depression on the BDI-SF. Sixty-seven (12.9%) of the 520 research participants disclosed suicidal thoughts within the last two weeks via the BDI-FS. None of these 67 participants were determined to be actively suicidal.
The distribution of characteristics of the total sample was similar across the methods of administration with the exception of age, last grade of education completed and depression. The prevalence of child physical abuse reported under each screening mode is presented in figure 1. Overall, 43.4% of participants disclosed childhood physical abuse. The face to face unstructured interview identified higher percentages of abuse (66.3%), followed by face to face structured screen (45.4%), ACASI (35.5%) and paper and pencil (35.1%) and was significantly different (p value <0.0001).
Childhood physical abuse was not associated with the selected covariates in the study population with the exception of depression ( Table 2). Participants who reported childhood physical abuse showed positive association with depression, with 31% depression in those who had physical abuse compared to 23% in those who did not (p=0.0380) We examined the effect of different screening modes to identify child physical abuse, our multivariable model adjusting for potential confounders showed that in both types of faceface interviews, the participants were more likely to report abuse. Specifically in complete case models, the estimated odds of identifying (being reported) child physical abuse is 4.3 times more in the unstructured face-to-face interviews with more probes compared to the paper and pencil screens, as shown in table 3. Similarly, the estimated odds of child physical abuse reported under structured interview was 1.5 times more compared to paper and pencil screen though the effect was not statistically different. Multiple imputation results show similar results to the complete case but the confidence intervals were a bit tighter interval and the structured face to face interview shows a marginal significance.

Discussion
The prevalence of childhood physical abuse identified by the face to face unstructured interview was four and one half times that of paper and pencil screen, significantly more than all three structured modes of administration. The interviewer who conducted the face to face interviews was a very experienced physician with an expertise in childhood abuse assessment which may account for some of this difference. Another possible contributor is the fact that the face to face unstructured interview allows further probing.
Only one prior study, by DiLillo and colleagues, has compared different modes of administration of screens to identify a history of both childhood physical abuse, comparing three modes (paper and pencil questionnaire, computer assisted survey and face to face structured interview), in a sample of female college students. 33 The DeLillo study reported an overall prevalence of childhood physical abuse of 15.5% but concluded that the mode of administration was unrelated to disclosure of a history of childhood physical abuse (X 2 =1.1; p value = 0.58). The present study found prevalence that was more than twice that; 38.6% vs. 15.5%, despite the latter asking about a history of childhood abuse that occurred before age 18 years and the present study using age 17 years as the cutoff. The large difference in prevalence between the two studies when comparing the structured modes of screening is most likely to be explained by differences between the study populations: the former sampled female students, in a college setting; an overwhelmingly white and middle class group. In contrast, the present study sampled males and females ages 12-24 years, who were 53% Hispanic and 41% non-Hispanic black, recruited from an urban poor population.
The two studies used two different measures to identify physical abuse: The former used the CAMI and the present study used the CMIS-SF, but it is unlikely that the difference in the instruments used in each study accounts for the large difference in prevalence, as both measures utilize detailed and behaviorally specific questions which is considered to be the most effective type of screen. 42, 34 The inclusion of the face to face unstructured interview as a fourth mode of administration in the present study is likely to account for the fact that when looking at overall prevalence of childhood physical abuse in this study, we see a prevalence triple that of DiLillo (44.5% vs. 15.5%).
The present study has some limitations. The retrospective, self-report has been found in some history of childhood physical abuse research to be somewhat unreliable due to error in recall resulting in false positives and false negatives. 35 Some researchers suggest that official child protective service reports and self-report used together should be the gold standard 36 but, this is not practical for studies in most settings where official childhood abuse records are not available. More important, a significant proportion -perhaps even a majority -of childhood abuse cases go unreported, so studies using only verified reported cases are likely to undercount. 1,2,24 , Indeed, a number of studies have shown that retrospective self-report has shown high stability over time. 37 Having one sole clinician for the administration of the unstructured interview rather than a number of clinicians with different levels of experience and comfort, an approach taken to reduce the influence of clinician variability on disclosure, limits the generalizability of the findings.

Conclusion
Although research on how best to identify childhood physical abuse history is in its infancy, this study suggests that face to face methods may offer the most effective ways to screen young people in primary care settings. However, because health care providers are not routinely inquiring about it, we need to better understand the trajectory from suspicion of abuse to the reporting of it in the primary care setting 18.19 . While the present study tells us nothing about how health provider training, experience, competency and comfort level influence the willingness to inquire about abuse, it does underline the need for further lines of research inquiry.
The effectiveness of a given mode of administration of screens to identify childhood abuse should not be confused with its practical application in the clinical setting. Health care providers in primary care practice settings face significant time pressures 38 and therefore, we need to examine whether face to face modes are the most labor intensive and time consuming compared to computer and paper or pencil questionnaires. Furthermore although computer technology is increasingly shaping health care, it is unclear how we will see the adoption of computer based screening for a range of health issues. 39 Computer based methods for communication between patient and health care provider still present significant challenges for primary care settings where they are not yet seen as practical. 40 Finding the screening method to identify childhood abuse that will prove to be most practical in the primary health care environment, where the use of technology is ever evolving, is a complex issue. Which mode of screening is most practical in the health care setting remains an open question.  Table 1 Distribution of study participant characteristics of study participants' by modes of administration of screens to identify childhood abuse:    Table 3 Adjusted odds ratios, 95% confident intervals and P value of the relationship of mode of administration of screens to identify childhood physical abuse and proportion of childhood abuse: