Using Routinely Collected Hospital Data for Child Maltreatment Surveillance: Issues, Methods and Patterns

In Australia and many other countries, data on child maltreatment are largely derived from child protection agencies, predominantly reporting on events reported for investigation and/or substantiated cases of child maltreatment [1]. This approach underestimates the incidence of maltreatment [2, 3] and makes inter-jurisdictional comparisons difficult, as each jurisdiction can have their own definition of what constitutes child maltreatment, unique definitions on what constitutes a report, and different processes for investigating and substantiating or refuting such reports. The World Health Organisation (WHO) in collaboration with the International Society for Prevention of Child Abuse and Neglect (ISPCAN) has called for common conceptual and operational definitions of child maltreatment to enable identification of cases across sectors involved in child maltreatment response and prevention [4]WHO and ISPCAN define child maltreatment as "all forms of physical and/or emotional ill-treatment, sexual abuse, neglect or negligent treatment or commercial or other exploitation, resulting in actual or potential harm to the child's health, survival, development or dignity in the context of a relationship of responsibility, trust or power" ([4], pg 9), with the four types grouped as physical abuse, sexual abuse, emotional or psychological abuse, and neglect.

Nationally and internationally, there has been a growing recognition of the importance of health professionals in identifying, documenting and reporting suspected child maltreatment [5, 6] and for child maltreatment to be recognised as a public health problem [7]. The WHO in collaboration with the United Nations Children's Fund (UNICEF) called for child maltreatment to be recognised as a global public health concern, with an urgent recommendation for uniform reporting procedures to register both fatal and non-fatal child maltreatment [5]. The WHO highlighted health professionals as being in the best position of all professions to obtain evidence of child maltreatment, and called for better systems to enable communication between health professionals and social services.

If health professionals identified, documented and reported suspected child maltreatment routinely in patients medical records, statistical data derived from these records could potentially provide information about populations at risk as well as populations where maltreatment is already present (as provided by data derived from child protection agencies). To use hospital morbidity data for surveillance of child maltreatment, clear operational case definitions are required to ensure the system is both sensitive and specific to ensure accurate identification of true cases and limited inclusion of false cases [4].

The information contained in the medical record is coded according to the International Classification of Diseases and Related Health Problems (ICD) system developed by the WHO [8], which provides an internationally standardised system for classifying and aggregating diseases, injuries, causes of injuries and related health conditions for statistical purposes. To provide greater specificity for morbidity data collection, many countries modify the international ICD for clinical purposes. The WHO sanctions these clinical versions, with all of these versions of the classification required to be comparable to the international version at the three/four character code level, though supplemental characters may be added to the base three/four character code to increase specificity for clinical purposes. In Australia as well as many other countries, including New Zealand, Ireland, Germany, Romania, Slovenia and Saudi Arabia[8], the clinical modification that is used is the ICD-10 Australian Modification (ICD-10-AM) ([9].

As the ICD is a statistical classification, with specificity of case identification a critical element, the ICD provides stringent rules which coders are required to follow when assigning ICD codes to medical records. For a clinical coder to apply a definitive maltreatment code ('T74 Maltreatment syndromes', 'Y06 Neglect and abandonment' or 'Y07 Other maltreatment syndromes'), there must be clear clinical documentation of evidence of maltreatment. If the presentation is injury related, the coder is first directed to assign a code/s to describe the nature of the injury itself or if the presentation is disease related, the nature of the disease and then a code/s to describe the external cause of the injury. A single external cause code includes three dimensions: intent (i.e. accidental, assault, or undetermined), mechanism (i.e. struck by, fall/push, poisoning etc), and object/substance involved (i.e. knife, firearm, etc)). If documentation in the medical record indicates that the cause of the injury/disease is 'queried' or 'suspicious' of maltreatment but evidence of further investigation to rule out maltreatment or to substantiate it is not documented, the coder cannot assign a definitive maltreatment code. Instead the coder may assign a range of codes indicating possible maltreatment (such as 'Z04.4 Examination and observation following alleged rape and seduction') or problems related to previous alleged maltreatment ('Z61.4 Problems related to alleged sexual abuse of child by person within primary support group') [9].

An outcome of these rigid coding requirements is that case identification using morbidity data is likely to be specific (i.e. limiting inclusion of false cases), but not sensitive (i.e. not identifying all true cases). Hence morbidity data are likely to underestimate the true magnitude of child maltreatment if the clinical documentation is not clear, concise and complete and unless a broad range of indicative codes are used to identify possible cases of child maltreatment. Schnitzer et al found that ICD-9-CM codes were highly specific indicators of maltreatment, with 100% of records containing documentation about maltreatment, almost 90% regarding current maltreatment and just over 10% describing a history of maltreatment [10]. However, Winn et al found 25% of injuries to children resulting from violence identified through a multi-hospital surveillance system were not coded as assault-related using ICD external cause codes and concluded that assault-related external cause codes were specific (99.7% specificity) but not sensitive (74.6% sensitivity) [11]. Recognising that definitive maltreatment codes are likely to miss true maltreatment cases, a variety of diagnosis and external cause ICD codes have been used in an attempt to better identify cases of child maltreatment in health data sets [12]. Recent research conducted using linked hospital data and child protection data for a prospective case-control birth cohort found risks of allegations of child maltreatment and risks of substantiations of child maltreatment increased by 1.49 and 1.74 respectively for each single hospital admission per year [13]. Furthermore, cases with a discharge diagnosis of a mental or behavioural disorder had a 26 times greater risk of a child maltreatment substantiation and cases with a discharge diagnosis of an injury or poisoning had a 21 times greater risk of a child maltreatment substantiation compared to other discharge diagnoses [13].

Even in the absence of population-based linked databases, hospital morbidity data collections could enable an efficient population-based nationally standardised source of data to provide information on child maltreatment-related separations and at-risk populations. However, this would rely on clear documentation and more inclusive coding to enable the coding of both cases of queried as well as confirmed maltreatment (whilst still clearly differentiation such cases). However, to date, there has been no research conducted which explores the issues, methods and patterns of child maltreatment information in routinely collected health data collections in the Australian context and only limited research, using previous versions of the classification system, at an international level. The aim of this paper is to stimulate discussion and contribute to the future development of non-fatal child maltreatment surveillance initiatives. This study aimed to describe the issues around case identification using ICD coded morbidity data, outline methods for selecting and grouping relevant codes, and illustrate patterns of maltreatment identified when using the outlined approach.

This study involved a systematic review of the ICD-10-AM classification system underpinning hospitalisation data to identify relevant codes for case identification of maltreatment-related presentations and a descriptive analysis of these data to explore patterns of code utilisation.

This study undertook a review of routinely collected and coded hospital data as a source of information on child maltreatment. While less than 0.5% of children overall hospitalised in Australia between 2005 and 2006 had a maltreatment code assigned in their separation data, almost 4% of children with a principal diagnosis of a mental and behavioural disorder and over 1% of children with an injury or poisoning as the principal diagnosis had a maltreatment code assigned. This result supports previous research which identified that cases with a discharge diagnosis of a mental or behavioural disorder or an injury or poisoning had significantly increased risk of child maltreatment substantiation using linked hospital and child protection data [13]^.

While the ICD-10-AM classification system provides both definitive codes for child maltreatment (such as T74 Maltreatment Syndrome) as well as possible codes for maltreatment (such as those indicating examination for or problems related to previous alleged maltreatment), in the current data, only 18.8% of cases identified as having any of the range of maltreatment codes were coded using the definitive T74 Maltreatment Syndrome code. Relying on this code alone would have significantly underestimated the prevalence of maltreatment-related presentations, with previous research indicating that definitive maltreatment codes are highly specific but not sensitive indicators of maltreatment [10, 11]. The codes more commonly assigned for those cases identified as having a maltreatment code were those referring to problems related to the alleged sexual abuse (27.4% Z61.4 and Z61.5) or alleged physical abuse (10.8% Z61.6).

The most commonly assigned principal diagnosis code in those children coded with a maltreatment code varied according to age group. In those younger than 5 years, the 'T74 Maltreatment syndrome' case was the most commonly assigned principal diagnosis code for children under 5 years of age', similar to O'Donnell et al who found that most children with a maltreatment code were aged under 5 years [16]. For older children the most common principal diagnosis for 6-9 year olds was 'F91 Conduct disorder', in the 10-14 year old age group a mixture of injury and mental and behavioural disorder codes, and for 15-17 year olds were also mental and behavioural disorders. A higher proportion of mental and behavioural disorder codes and injury codes for children who are known to the child protection system has previously been identified by researchers in Western Australia [13]. Patterns of maltreatment types varied by sex and age group. Physical abuse was most common overall for males with 62.6% of males with a maltreatment code assigned a physical abuse code. Sexual abuse was most commonly assigned overall for females with 52.9% of females with a maltreatment code assigned a sexual abuse code. Similarly, O'Donnell et al found a much higher proportion of females (71.4%) than males (28.6%) aged between 2-13 years with a notified sexually transmitted infection [16]. However, in females younger than 1 year of age, the largest proportion of cases with a maltreatment code were assigned an 'Other abuse' code (37.9%), and in females aged 1-5 and 6-9 females, the largest proportion of cases were assigned a 'Physical Abuse' code (36.2% and 55.6% respectively).

For cases with a maltreatment code and a perpetrator code, the most common perpetrator coded for all age groups up to 9 years was parent, carer or family member. However, within these age groups there were 7 cases coded as having a perpetrator of 'spouse', which is defined as the spouse of the injured patient. It is possible that these cases were cases of child maltreatment by the 'spouse' of the parent (i.e. step-parent or defacto partner of parent) which were wrongly assigned a 'spouse' code. A conservative estimate of maltreatment-relatedness of cases was used in this study, through removing all other and unspecified perpetrator codes for children aged 15-17 years of age. This may have underestimated the number of cases in this age group who were victims of child maltreatment, and further research is needed to examine this group in more detail to enable a better estimate of the likely proportion of child maltreatment cases for 15-17 year olds. However, recent research has found that around 50% of interpersonal violence cases (all ages) are assigned an 'other' or 'unspecified' perpetrator code [17].

Patterns of diagnoses by maltreatment codes/maltreatment types can provide valuable insight into the epidemiology of severe child maltreatment and the resultant harms to children over time. Were researchers able to have some confidence in the sensitivity and specificity of maltreatment codes, these data could form a valuable source of information to target intervention, reduction and prevention initiatives.

A potential limitation of this study, and any study reliant on the clinical coding of morbidity data, is the requirement for complete and accurate clinical documentation and consistent, quality clinical coding to ensure reliable coded data. The capture of complete, timely, relevant and accurate maltreatment data is reliant on a number of processes. Firstly there needs to be a suspicion of child maltreatment in the clinical context, the clinical documentation of indicators of maltreatment for children suspected of maltreatment needs to be provided and the coder needs to assign relevant ICD codes to identify the type of maltreatment documented. If the documentation in the medical record is ambiguous, incomplete or illegible, coders are unable to assign appropriate codes. This could result in underestimations of child maltreatment in coded morbidity data. In order to quantify this underestimation a medical record and coding audit is required, and further stages of our current research study seek to explore these issues.

This study has demonstrated that hospital morbidity data could provide valuable information for routine monitoring and surveillance of non-fatal child maltreatment even in the absence of population-based linked data sources. ICD coded hospital morbidity data provides codes for both definitive and possible maltreatment as well as types of maltreatment, and being a statistical classification, there are strict rules and definitions to improve the specificity of case identification. Further work is needed to evaluate the reliability of coding of identified maltreatment codes and to enumerate the likely sensitivity and specificity of the system for case identification of maltreatment-related presentations. In the meantime, health professionals and administrators can use this information to conduct follow-up audits of possible cases of maltreatment. Where linked databases are available, researchers could use these codes to further interrogate hospitalisation data for indicators of maltreatment in addition to the use of disease diagnosis codes and child protection outcome data. With national and international calls for a public health response to child maltreatment, better understanding of, investment in and utilisation of our core national routinely collected data sources will enhance the evidence-base needed to support an appropriate response to children at risk.