Daily Tackling Heath Inequalities: French Child Individual-Level Deprivation Index Development and Validation

The purpose of this study was to develop and validate a pediatric individual-level index for deprivation, usable in clinical practice and in public health. The index had a 4 phases development: items generation with literature review and experts interviews, items reduction with steering committee consensus, index derivation with multivariate analysis, and index validation with psychometric and Pearson analysis. French Child Individual-Level Deprivation Index (FrenChILD-Index) was addressed by untrained healthcare professionals in a cross-sectional multicentric study. The deprivation burden was blindly evaluated in every domain of lifestyle by an expert. Children in need of one specic type of healthcare for deprived children were: moderately deprived. Children in need of referral to a socio-medical unit for access to healthcare were: severely deprived. The main outcome measure was the agreement between FrenChILD-Index results and expert evaluation. Development phases produced a 12-item instrument. Validation phases were carried out in a 986 children sample. FrenChILD-Index fullled the Terwee validity criterion for screening instruments. For moderate deprivation, sensitivity was 96.0% [92.6; 98.7] and specicity 68.3% [65.2; 71.4]. For severe deprivation, sensitivity was 96.3% [92.7; 100] and specicity 91.1% [89.2; 92.9]. It correlated with the number of lifestyle deprived domains 0.80 [0.77; 0.83] and the amount of specic healthcare for children 0.86 [0.83; 0.88]. Conclusions: FrenChILD-Index is the rst pediatric individual-level index of deprivation, methodologically validated in Europe. FrenChILD-Index enables individual appropriate referral for deprived children. It enables considering social determinants of health into account in epidemiological adjustment, patient sample stratication and program impact measurement.


Introduction
Social determinants of health have a tremendous impact on children, generating early health inequities [1][2][3], reinforced by further barriers in healthcare access [4]. Deprivation is known to be multidimensional, affecting every domain of living conditions: economic stability, but also education, health and healthcare, neighborhood and built environment (including housing), social and community context, and family context [5][6][7][8].
Area-based deprivation indexes have been used in Europe for children in epidemiological studies [1,2,4].
But this kind of index has already shown some limitations with signi cant discrepancies [9]. Furthermore they are less appropriate in clinical practice to tackle health inequalities in adapting individual care.
It is common for most pediatricians, regardless of their clinical practice settings, to provide care to deprived children, given the signi cant prevalence of economically poor children (21.0% in France, 2018) [12]. Screening for deprivation is crucial to provide effective interventions for children [13], but pediatricians have to know how to proceed [14]. However, to our knowledge, no convenient individual-level index for child deprivation has been validated in France or in Europe.
Deprived children need several kinds of social support and healthcare [10]. In France, speci c Socio-Medical Units For Access to Healthcare (SMUFAH) are located in hospitals for severely deprived patients with complex needs. These units offer several speci c types of healthcare: social assessment to gain access to health insurance, free medical consultations, free medicine dispensing, physical accompaniment in healthcare, home visit because of unhealthy housing, and multidisciplinary care coordination meetings. The SMUFAH multidisciplinary care plans combine at least two of these speci c types of healthcare for deprived children [11].
The aim of this study was to develop and validate a pediatric individual-level index for deprivation, usable in clinical practice and public health: French Child Individual-Level Deprivation Index (FrenChILD-Index).

Methods
The study design included four phases: item generation, item reduction and writing, index derivation, and index validation.
For the item generation phase, a literature review was performed in October, 2013 to extract existing screening instruments for investigating deprivation, vulnerability or poverty at individual level, in PubMed (for main international references) and CAIRN (increasing exhaustiveness to articles in French). In addition, face-to-face semi-structured interviews with 13 senior experts in speci c healthcare for deprived children collected their opinions on the domains to be measured, and on the complementary items to address. Interviews ended when data saturation was achieved.
For the item reduction and writing phase, a steering committee was set up. Eight of the 13 senior experts accepted to participate (one pediatrician, two nurses, two social workers, two healthcare and socioeducational managers, and one health mediator). They selected by oral consensus, relevant and acceptable items covering each dimension of social deprivation, proposed their wording and an informative appendix.
An independent reading group (one general practitioner, two pediatricians, three nurses and two social workers) validated the choice and wording of the items and appendix, via an online questionnaire with a semi-quantitative notation (from 1: total disagreement to 9: total agreement) of the wording and relevance of each question. Their agreement about appropriateness was calculated according to analysis rules [15].
For the index derivation and validation phases, 13 items judged as being appropriate or uncertain were tested, by healthcare professionals untrained in their use, on a sample of children. This sample was derived form a cross-sectional multicentric study conducted between April 2018 and October 2019, which recruited a convenient sample of children in two French university hospitals (Marseille, Nice). Children were eligible when aged 3 to 15 years old, admitted to paediatric emergency units without life-threatening conditions or ongoing medico-legal procedures. Children could only be selected once. Only one sibling was included. Informed consent was obtained from the participant's legally authorized representative and the child him or herself when 8 years old or above. For non-French speaking children, a professional telephone interpretation service was used (ISO 13611:2014;17100:2015). A sample size about 1,000 children was required to show a sensitivity and speci city of 80% with a 95% con dence interval (CI) and a precision of 5% for primary outcome, considering an expected prevalence of deprivation for at least 25% of the sample (one hospital being located in a deprived area) [12,16].
The primary outcome of whole deprivation burden assessment was based on a blinded expert evaluation for each child. This expert was a trained healthcare professional working in a SMUFAH (social worker, nurse or pediatrician; with at least one year of experience caring for deprived children). He/she assessed the whole deprivation burden with several criteria: (1) the type and number of deprivation domains by following the concept of deprivation, in every domain of lifestyle at an individual level [5,6], (2) the type and amount of speci c healthcare required for deprived children (both lists mentioned in the introduction), and (3) the need for admission to SMUFAH (when at least two different types of speci c healthcare for deprived children were needed) [11].
For the index derivation phase, item-internal consistency was assessed by correlating each FrenChILD-Index item with the deprivation domain they were logically related to. Items were retained if they had a signi cant moderate correlation (Pearson correlation coe cient |r| ≥ 0.3). Internal consistency was assessed using Cronbach's alpha coe cient. We performed two linear multiple regressions to test retained items for predicting the whole deprivation burden with: i) the number of deprivation domains affected and 2) the amount of speci c healthcare needed for deprived children. For the FrenChILD-Index scoring, items were weighted on the standardized coe cients (average of the two regressions) and revised according to expert recommendations.
For the index validation phase, discriminant properties were assessed accordingly: i) at least one speci c type of healthcare for deprived children = moderate deprivation; and ii) at least two speci c types of healthcare for highly deprived children = severe deprivation, to be admitted to SMUFAH. Sensitivity and speci city were calculated with 95% con dence intervals.
FrenChILD-Index reproducibility was assessed via a phone call retest on a random sample (0.3%) of children after 5 to 6 months (this period was longer than schooling or health insurance applications if they had already been started). It included children in a situation perceived as stable by parents and not admitted to SMUFAH over the period and was strati ed with no more than 1/3 of children without any deprivation criteria (initial FrenChILD-Index = 0).
Complete-case analysis was drawn. Statistical analysis used chi-squared, Fisher's and Student's tests, Pearson correlations, linear multiple regressions and ROC curve analyses with SPSS 20.0 (IBM, Armonk, NY, USA) and SAS 9.4 (SAS, Cary, NC, USA) softwares. Bootstrap con dence intervals were calculated using 1,000 samples generated by unrestricted random sampling with strati cation according to expert assessment (mean sample size = 624). Signi cance threshold was 0.05.

Results
This study shows the development of the FrenChILD-Index and its validation by resampling following the TRIPOD guidelines [17]. Study ow chart is presented, Fig. 1.
In the item generation phase, 52 different items were extracted from 7 published indexes [18-24] and 13 senior expert interviews. Two indexes were excluded because one concerned child health vulnerability [25], and the other used undirected interviews and validity wasn't assessed [26]. Item generation and reduction are presented in Supplement 1.
In the item reduction phase, 13 items were selected by the steering committee. Items needed to: cover the six deprivation domains listed above; be understandable (included for cases of low literacy); help to develop a care plan, obtain social support or improve family habits; and be acceptable in pediatric care (simple closed and easy to translate questions). The steering committee added an appendix for professionals, so that they understood the questions in the same way, and had elements for item justi cation and appropriate referral for healthcare and social support.
The reading group reordered the items, and judged 10 of the 13 items to be appropriate and 3 uncertain. There was no inappropriate item.
In the index derivation phase, the cross-sectional multicentric study included 990 children. Their characteristics are presented in Table 1. Item derivation including univariate and multivariate analysis is presented in Supplement 2.
Twelve items were included in multiple linear regressions. The household vulnerability item was dropped because the reading group judged its appropriateness uncertain and it had no relevant correlation.
Senior experts corrected FrenChILD items weighting as little as possible to ful ll several logical conditions: (1) non-French speaking had an independent and long-lasting impact (recent migration weight was shared with non-French speaking) [27]; (2) single parenthood alone did not require assessment by a social worker; (3) homeless children or with incomplete health insurance and not knowing any social workers should be admitted to SMUFAH [11]; and (4) homelessness was a wider determinant of health than lack of health insurance (homelessness weight had to be higher). The frequency of deprivation domains per FrenChILD-Index level is presented in Fig. 2. It was correlated with each speci c type of healthcare for deprived children (all r ≥ 0.32).
The impact of FrenChILD-Index on children pathway was measured using psychometric properties. A FrenChILD-Index retest for 31 children was highly correlated (r = 0.78; p < 0.01), the mean difference was low (-0.71; SD = 7.51).

Content validity
The objectives, target population and concept measured were clearly described and senior experts were involved in item selection. FrenChILD-Index items signi cantly correlated with each corresponding domain of deprivation.

Internal consistency
Items showed a good correlation with each other. Their weighting did not use factorial analysis proposed by Terwee. This method resulted in weightings that were not strongly enough correlated with speci c healthcare needs for deprived children and deprivation dimensions, both amounts taken as references for progressivity.

Criterion validity
Blinded cross-evaluation by an expert was already used by Pascal and Colvin [23,29]. In clinical practice, experts are rarely available. Indeed the aim of FrenChILD-Index is to replace them.

Construct validity
FrenChILD-Index showed how deprivation domains accumulate to threaten the appearance of early health inequalities [6,7]. Each item signi cantly correlated with its deprivation domain. Most were also correlated with the number of deprivation domains and amount of speci c healthcare for deprived children.

Individual interpretability
FrenChILD-Index showed an excellent sensitivity and speci city for each operational cut-off (moderate ≥ 6 and severe deprivation ≥ 26). This allows e cient child referral (Fig. 3, Supp3). The appendix provided the additional information for appropriate referral. Lack of information for appropriate referral was known to be a barrier to screening [10,30].

Collective interpretability
FrenChILD-Index scaling has a su cient number of degrees for describing various deprivation levels in a sample of children. It will enable quantitative monitoring in clinical and epidemiological research but also in actions to reduce social determinants of health. It is useful for studying early health inequalities by grouping all of the deprivation domains in a linear proportionality.
Reliability was good in the retest evaluation. This highlights the stability of FrenChILD-Index over time.
A oor but no ceiling effect, FrenChILD-Index has a oor level because it focuses on deprived children. Having no ceiling effect, it is able to describe a wide spectrum of deprivation levels.
Sokol reviewed 11 indexes derived in the USA in children but only 3 were validity assessed [5]. IHELP had a similar sensitivity and speci city in the USA [29]. PSQ was out of scope because questioning and screening concerned parental depression [31]. Six indexes questioned adverse childhood experiences. These were neither included in items nor retained as validation criteria in FrenChILD-Index. Senior experts considered that screening for deprivation (socially considered to be a condition for assistance) and addressing adverse childhood experiences (socially banned and legally prohibited) should be considered separately because the latter are deemed to be judgmental and cause response biases [32]. WHO also distinguishes between social determinants of health assessment and reduction, and adverse childhood experiences prevention, in separate chapters of its strategy [33].
In France, Fouchard compared three adult individual-level deprivation indexes [34]. EPICES tool had the most common points with FrenChILD-Index but not all of the items were appropriate for children [22]. Furthermore, the interpretation of EPICES was expected to be more perturbed by lifestyle confounding factors than FrenChILD-Index because of items on leisure activities (e.g. as highlighted in case of access restriction).
Use of FrenChILD-Index will improve deprivation screening in clinical practice and as Garg showed, systematic screening increases appropriate referrals by more than 4 [35].
FrenChILD-Index highlighted an expert judgment bias. Experts reported less family context deprivation in cases of homelessness but not in cases of interculturality. Indeed, even for experts, homelessness raises questions about the lack of standards on how to be a good enough parent in such severely deprived living conditions. This study had one limitation. FrenChILD-Index identi ed 172 severely deprived children, whereas experts retained referral to SMUFAH for only 93. SMUFAH is a relevant resource for access to healthcare. Indeed, this highlights the wider support needed for those families. However, severely deprived children suffered from deprivation in several domains and other types of social support. Economically poor children were expected to be much more frequent in emergency units [12]. Furthermore, expert evaluation is already a valuable type of care in complex living conditions [10]. FrenChILD-Index only gave 3 (0.3%) false negatives.

Conclusion
We have developed and validated FrenChILD-Index, in response to the lack of any validated pediatric deprivation index in Europe. It is now used in France for appropriate individual referral and for epidemiological adjustment to deprivation biases and available for strati cation of children samples and assessment of impact in public health interventions. It will improve professional knowledge about social determinants of health and patients' early health inequities.