ICD-10 codes for surveillance of non-fatal abusive head trauma in Aotearoa New Zealand: a retrospective cohort study

Abstract

Objectives

To assess the validity of an International Classification of Diseases (ICD) code based definition of non-fatal head trauma caused by child abuse (abusive head trauma) for population surveillance in New Zealand.

Design

A retrospective cohort study of hospital inpatient records.

Setting

A tertiary children’s hospital in Auckland, New Zealand.

Participants

1731 children less than 5 years of age who were discharged after a non-fatal head trauma event over a 10-year period from 1 January 2010 to 31 December 2019.

Outcome measures

The outcome of assessment by the hospital’s multidisciplinary child protection team (CPT) was compared with the outcome of ICD, Tenth Revision (ICD-10) discharge coding for non-fatal abusive head trauma (AHT). The ICD-10 code definition of AHT was derived from an ICD, Ninth Revision, Clinical Modification definition developed by the Centers for Disease Control, Atlanta, Georgia, which requires both a clinical diagnosis code and a cause-of-injury code.

Results

There were 1755 head trauma events with 117 determined as AHT by the CPT. The ICD-10 code definition had a sensitivity of 66.7% (95% CI 57.4 to 75.1) and specificity of 99.8% (95% CI 99.5 to 100). There were only three false positives but 39 false negatives, with 18 of the false negatives coded with X59 (exposure to unspecified factor).

Conclusions

The ICD-10 code broad definition of AHT is a reasonable epidemiological tool for passive surveillance of AHT in New Zealand but it underestimates the incidence. Its performance could be improved by clear documentation of child protection conclusions in clinical notes, clarifying coding practice and removing the exclusion criteria from the definition.

STRENGTHS AND LIMITATIONS OF THIS STUDY.

• This study examines the sensitivity and specificity of International Classification of Diseases 10 discharge codes for the passive surveillance of paediatric abusive head trauma, compared with a gold standard.

• This study investigates a large cohort of head injured children over a 10-year period, using data from a single centre with consistent clinical and coding practice.

• This was a retrospective study, which meant that data could not be collected concerning the reasons behind, and timing of, discharge coding decisions.

• This was a study in a tertiary children’s hospital, so its findings are not necessarily generalisable to clinical and coding practices in smaller centres.

Introduction

Abusive head trauma (AHT) has become the accepted term for non-accidental head injury in children under 5 years of age, where that head injury involves skull fracture and/or intracranial injury.^{1 2} The single most common intracranial injury in AHT is subdural haemorrhage, but a wide variety of injuries all come under the diagnostic umbrella.³ This makes passive surveillance challenging as there is no single International Classification of Diseases (ICD) code that allows cases to be identified.

AHT has significant mortality (in New Zealand, around 20%)^{4 5} and many survivors experience serious long-term neurodevelopmental effects.^6–8 As a society, we should work to reduce the incidence of AHT, but in order to assess the effectiveness of interventions we need a simple, reliable way of measuring the incidence.

The need for better population surveillance of AHT has been recognised for some time.^{9 10} Between 2008 and 2011, a CDC expert panel developed definitions for surveillance using combinations of ICD, Ninth Revision, Clinical Modification (ICD-9-CM) codes in children under 5 years of age.¹¹ The ICD-9-CM was at that time ‘the official system of assigning codes to diagnoses and procedures associated with hospital utilisation in the US’.¹² Narrow and broad definitions were created with the broad definition being recommended for population-based surveillance. The broad definition required either the ‘shaken baby syndrome’ code, 995.55 (absent from versions of the ICD used outside the US) or a head injury code (such as intracranial injury) combined with an external cause-of-injury code for either assault or undetermined intent. This broad definition has been used in multiple studies of the incidence of non-fatal AHT in the US^13–18 and one in Canada, where it was transposed into equivalent codes in the International Statistical Classification of Diseases and Related Health Problems, 10th Revision, Canada (ICD-10-CA).¹⁹ The ICD-9-CM broad definition was found to have a high sensitivity and specificity for AHT diagnosed by a tertiary hospital-based child protection team (CPT) in Pittsburgh.²⁰

In New Zealand, there is no passive surveillance of AHT. Estimates of the incidence have been derived by combining the results of retrospective cohort studies with one prospective national survey twenty years ago.^{3 21} It is not feasible to conduct such studies repeatedly. However, all hospitalisations are coded using the International Statistical Classification of Diseases and Related Health Problems, Australian Modification (ICD-10-AM).²² The objective of this study was to examine the validity of the CDC broad definition, adapted for use with the ICD-10-AM, for passive surveillance of non-fatal AHT in New Zealand.

We did not examine the validity of the CDC definition for fatal AHT, as that definition is based on death certificate data and would require a separate, national study.²³

Methods

Setting

The study was conducted in Starship Children’s Health, Auckland, New Zealand. Publicly funded healthcare in New Zealand is administered through multiple districts. Starship is the national tertiary paediatric trauma centre, with the only specialist paediatric neurosurgical and intensive care services. District hospitals admit children with head injuries nationwide, and only those requiring tertiary care are transferred to Starship. However, in metropolitan Auckland (three districts) and Northland (north of Auckland), most children with intracranial injuries are transferred to Starship. In the 2018 census, Auckland and Northland had a combined population of 1.75 million, 37% of the national population of 4.7 million.²⁴

Starship Children’s Health has a multidisciplinary CPT, known as Te Puaruruhau. Starship clinical guidelines require any clinicians with concern about possible abuse to consult Te Puaruruhau. If agreed that referral to the CPT is appropriate, a detailed assessment takes place. Once all relevant information has been gathered, an overall conclusion as to the likelihood of AHT is reached, in collaboration with the statutory authorities. The process (and its relationship to statutory care and protection and prosecution outcomes)^{5 25} is described in detail elsewhere,²⁶ and is recognised in the literature as ‘a thorough clinical and forensic investigation’.²⁷ There are three possible conclusions: AHT, undetermined or other (accidental injury or natural causes). Undetermined cases receive a full multi-agency investigation and are often concluded to be children at risk, even if the cause of the head injury cannot be established with certainty.²⁷

Hospital inpatient notes are mainly on paper, reviewed after discharge by clinical coders who assign ICD codes, which are uploaded electronically to the Ministry of Health. One admission may include multiple consecutive events coded as separate ‘discharge events’.

Study design and population

ICD-9-CM codes in the CDC broad definition of non-fatal AHT¹¹ were transposed into equivalent ICD-10-AM codes as in the Canadian study.¹⁹ ICD-10 does not have an equivalent for the shaken baby syndrome code (995.55) used in the USA. The definition requires both a clinical diagnosis code and a cause-of-injury code. Clinical diagnosis codes include intracranial injury, skull fracture, crushing injury of head, multiple other injuries of head and concussion. The broad definition includes both ‘presumptive’ cause-of-injury codes (assault and maltreatment codes) and ‘probable’ cause-of-injury codes (undetermined intent). Those with codes for birth trauma, sexual abuse, coagulation defects, unintentional gun-related injury and congenital malformations are excluded from the definition.

A 10-year period was chosen for this study to provide a robust sample size. The Starship Business Intelligence team electronically provided ICD-10 codes and demographic data from 1 January 2010 to 31 December 2019 for all non-fatal discharge events of children less than 5 years of age with the head trauma diagnosis codes included in the AHT criteria (with or without any of the cause-of-injury codes).

Patient involvement statement

Patients or the public were not involved in the development of the research question, the design of the study or in recruitment to or conduct of the study

Data collection

Demographic information was collected included age, gender, resident district, ethnicity and deprivation decile. Ethnicity is recorded by self-report using a standardised prioritised system as follows: Māori, Pacific Islander, Asian, other and then European.²⁸ Deprivation decile is derived from the socioeconomic deprivation index (NZDep).²⁹ Deprivation scores are allocated to ten deciles, with decile 1 representing the least deprived. The demographics of the patient population included in this study are provided in table 1.

Table 1.

Demographics of 1754 non-fatal head trauma cases under 5 years of age that were coded with any of the ICD-10 diagnosis codes required in part for the coding-based classification of non-fatal AHT*

		n (%)
Gender	Female	777 (44.3)
	Male	977 (55.7)
Age (years)	<1	612 (34.9)
	1	408 (23.3)
	2	303 (17.3)
	3	211 (12.0)
	4	220 (12.5)
Ethnicity	European	830 (47.3)
	Māori	331 (18.9)
	Pacific Peoples	302 (17.2)
	Asian	257 (14.7)
	Others	34 (1.9)
Deprivation decile†	1–3	476 (27.4)
	4–7	614 (35.3)
	8–10	650 (37.4)
Home region†	Metropolitan Auckland+Northland	1578 (90.5)
	Other North Island	144 (8.3)
	South Island	21 (1.2)
Time period	2010–2014	910 (51.9)
	2015–2019	844 (48.1)

*Coding criteria for non-fatal AHT requires any of the head trauma diagnosis codes in combination with any of the cause-of-injury codes and without any of the exclusion codes.

†The home region was unrecorded for 11 cases and deprivation unrecorded for 14 cases.

AHT, abusive head trauma; ICD-10, International Classification of Diseases, Tenth Revision.

There were 1821 discharge events, representing 1754 cases of head trauma involving 1731 children (21 children had 2 episodes of head trauma and 1 child had 3 episodes of head trauma). The codes for the 1821 discharge events are presented in table 2. Of 1754 head injury cases, 84 had one or more required cause-of-injury codes. Three of the 84 had exclusion codes, leaving 81 cases that met ICD-10 cause-of-injury code criteria for non-fatal AHT. The diagnosis codes for the 81 AHT cases are provided in table 2 and their cause-of-injury codes in table 3.

Table 2.

ICD-10 non-fatal AHT diagnosis codes and their distribution in all 1821 hospital events for 1754 cases under 5 years of age with any of those codes*, and in the 81 of those cases that met the full ICD-10 coding criteria for AHT†

ICD-10 code and description		All hospital coded head trauma events (N=1821)	Head trauma cases which met AHT coding criteria (N=81)
XIX. Injury, poisoning and certain other consequences of external causes		n (%)	n (%)
S02: fracture of skull and facial bones		475 (26.1)	22 (27.2)
S02.0	Fracture of vault of skull	350 (19.2)	19 (23.5)
S02.1	Fracture of base of skull	165 (9.1)	3 (3.7)
S02.7	Multiple fractures involving skull and facial bones	0 (0.0)	0 (0.0)
S02.8	Fractures of other skull and facial bones	20 (1.1)	1 (1.2)
S02.9	Fracture of skull and facial bones, part unspecified	2 (0.1)	0 (0.0)
S06: intracranial injury		1592 (87.4)	77 (95.1)
S06.0	Concussion	1203 (66.1)	13 (16.0)
S06.1	Traumatic cerebral oedema	11 (0.6)	3 (3.7)
S06.2	Diffuse brain injury	43 (2.4)	9 (11.1)
S06.3	Focal brain injury	73 (4.0)	9 (11.1)
S06.4	Epidural haemorrhage	93 (5.1)	3 (3.7)
S06.5	Traumatic subdural haemorrhage	271 (14.9)	68 (84.0)
S06.6	Traumatic subarachnoid haemorrhage	99 (5.4)	20 (24.7)
S06.7	Intracranial injury with prolonged coma	0 (0.0)	0 (0.0)
S06.8	Other intracranial injuries	16 (0.9)	2 (2.5)
S06.9	Intracranial injury, unspecified	14 (0.8)	0 (0.0)
S07: crushing injury of head		0 (0.0)	0 (0.0)
S07.1	Crushing injury of skull	0 (0.0)	0 (0.0)
S07.8	Crushing injury of other parts of head	0 (0.0)	0 (0.0)
S07.9	Crushing injury of head, part unspecified	0 (0.0)	0 (0.0)
S09: other and unspecified injuries of head		30 (1.6)	0 (0.0)
S09.7	Multiple injuries of head	0 (0.0)	0 (0.0)
S09.8	Other specified injuries of head	30 (1.6)	0 (0.0)

*Events and cases can be coded with more than one head trauma diagnosis code.

†Coding criteria for non-fatal AHT requires any of the head trauma diagnosis codes in combination with any of the cause-of-injury codes and without any of the exclusion codes.

AHT, abusive head trauma; ICD-10, International Classification of Diseases, Tenth Revision.

Table 3.

ICD-10 non-fatal AHT cause-of-injury codes and their distribution in the 81 cases* that met the full ICD-10 coding criteria for AHT†

ICD-10 code and description		N (%)
T74: maltreatment syndromes		16 (19.8)
T74.1	Physical abuse	16 (19.8)
T74.8	Other maltreatment syndromes	0 (0.0)
T74.9	Maltreatment syndrome, unspecified	0 (0.0)
XX. External causes of morbidity and mortality		81 (100)
Assault
Y00	Assault by blunt object	1 (1.2)
Y01	Assault by pushing from high place	0 (0.0)
Y04	Assault by bodily force	11 (13.6)
Y07	Other maltreatment	48 (59.3)
Y08	Assault by other specified means	0 (0.0)
Y09	Assault by unspecified means	18 (22.2)
Event of undetermined intent
Y29	Contact with blunt object, undetermined intent	0 (0.0)
Y30	Falling, jumping or pushed from a high place, undetermined intent	2 (2.5)
Y33	Other specified events, undetermined intent	0 (0.0)
Y34	Unspecified event, undetermined intent	2 (2.5)

*Cases can be coded with more than one of the cause-of-injury codes.

†Coding criteria for non-fatal AHT requires any of the head trauma diagnosis codes in combination with any of the cause-of-injury codes and without any of the exclusion codes.

AHT, abusive head trauma; ICD-10, International Classification of Diseases, Tenth Revision.

To check for cases of AHT which may have been missed because they were not correctly coded as head trauma, we also reviewed all discharge events of children less than 5 years of age with one of the cause-of-injury codes in the definition, but none of the diagnosis codes. Of 44 such events identified, one case should have had a head trauma diagnosis code. Thus, in all there were 1755 distinct non-fatal head trauma cases identified.

Of the 1755 cases, 263 referrals to the CPT were identified by cross-matching to the CPT referral database and review of clinical notes. Review of the CPT referral database found no cases of AHT that had not been identified by the ICD code search. The CPT conclusion for those 263 cases was accidental injury or natural cause in 146, ‘definite’ AHT in 94 and ‘undetermined’ in 23. When the CDC developed its broad definition for population surveillance, head injuries with an external cause-of-injury code for ‘undetermined intent’ were classified as ‘probable’ AHT. This was because a child injured in such circumstances may well have sustained AHT, even if the degree of clinical certainty recorded was insufficient for coders to attach an assault or maltreatment code.¹¹ Following the same principle, in this study cases concluded to be ‘undetermined’ by the CPT were classified as probable AHT, acknowledging that ‘undetermined’ cases may well signify AHT.^{21 30} This gave 117 ‘gold standard’ cases classified as definite or probable AHT. The 1492 unreferred cases were combined with the 146 referred cases that were determined to be accidental or natural cause, for a total of 1638 ‘gold standard’ cases classified as not AHT.

Statistical analysis

Data were managed in MS Excel for MS 365 and analysed using JMP V.15 (SAS Inc.) and StatsDirect V.3.2.7 (StatsDirect). Case classifications by ICD-10 code search were compared with the gold standard determinations by the CPT using standard diagnostic test performance metrics (sensitivity, positive predictive value (PPV), specificity and negative predictive value (NPV)—see online supplemental figure for further explanation).

Cohen’s kappa agreement statistic was also calculated. Comparisons of diagnostic test performance metrics between subgroups of cases were undertaken using comparison tests of proportions. The coding data and clinical notes were scrutinised for all of the cases where the classification as AHT by ICD-10 code search was different from that by the CPT (ie, the false positives and false negatives).

The total and annual incidences of AHT in metropolitan Auckland and Northland were calculated for cases classified by ICD-10 code search and determined by the CPT, in infants (under 1 year of age) and in children under 5 years. For infants the denominator used was annual live births by district, and for children under 5 years, it was annual estimated population age of 0–4 years, by district.³¹ Poisson distribution based rate comparison and regression methods were used to compare incidences and calculate lines of best fit.

Results

Non-fatal head trauma cases classified as AHT or not by the ICD-10 code search were compared with the gold standard determination by the CPT (table 4). Overall, the agreement was ‘substantial’³² (κ: 0.78, 95% CI 0.71 to 0.84). Of the 81 head trauma cases classified as AHT by the ICD-10 code search, 78 were true positives and three were false positives (PPV: 96.3%, 95% CI 89.6 to 99.2). Of the 1674 head trauma cases not classified as AHT by the ICD-10 code search, 1635 were true negatives and 39 were false negatives (NPV: 97.7%, 95% CI 96.8 to 98.3). The sensitivity of the ICD-10 coding based screening was 66.7% (95% CI 57.4 to 75.1) and the specificity was 99.8% (95% CI 99.5 to 100).

Table 4.

Diagnostic test table for the ICD-10 code search classification as non-fatal AHT in children under 5 years of age compared with the CPT’s determination of those cases*

		CPT determination of AHT
		Present	Absent
ICD-10 code search classification as AHT	Positive	78 (66.7%)	3 (0.2%)	81
	Negative	39 (33.3%)	1635 (99.8%)	1674
		117	1638	1755

*AHT was considered absent for 1492 cases not referred for CPT assessment.

AHT, abusive head trauma; CPT, child protection team; ICD-10, International Classification of Diseases, Tenth Revision.

The three false positives included two children not referred to the CPT. Both sustained injuries as a result of the witnessed actions of other children. One received an undetermined intent cause-of-injury code, the other an assault code. The third child sustained an unusual injury involving a complex fall. Following CPT assessment, this injury was concluded to be accidental, but it was coded with an assault code.

The 39 false negatives comprised 23/94 head injury cases concluded to be definite AHT and 16/23 cases concluded to be undetermined by the CPT.

With regard to the 23 false negatives that had been concluded by the CPT to be definite AHT, three were coded with both an AHT diagnosis code and an AHT cause-of-injury code but not classified as AHT by ICD-10 code search because they also had an exclusion code (two coagulation defect codes and one congenital malformation code). One false negative was coded with an AHT cause-of-injury code but no head trauma code. The other 19 all lacked an ICD-10 AHT cause-of-injury code. In all 19, there was either no history of trauma, or a history of injury considered to be inconsistent with the injuries sustained. In one case, there was not a clear conclusion in the inpatient notes. In many cases, the clinical notes included phrases like ‘probable non-accidental injury’, which may explain why some were not coded as abuse. However, there were seven cases where ‘non-accidental injury’ and ‘child abuse’ were written in the notes but not coded as abuse.

As with those 19 definite cases, all 16 false negatives in undetermined AHT cases lacked an ICD-10 AHT cause-of-injury code. In 10/16, the final CPT conclusion was not documented in the inpatient notes but only in a later report.

In total, 11/19 definite cases and 7/16 undetermined cases without an ICD-10 AHT cause-of-injury code were coded with X59 (exposure to unspecified factor). The occurrence of the X59 code among false negatives (18/35, 51%) was significantly greater than its occurrence among true negatives (11/1635, 0.7%, p<0.001).

Diagnostic test performance metrics did not significantly differ between cases based on their ethnicity, age, deprivation scale or year of admission (data not shown).

The combined 2010–2019 incidence for non-fatal AHT in metropolitan Auckland and Northland, as determined by the CPT, was 27.0 (95% CI 20.9 to 34.4) per 100 000 per year for infants and 6.4 (95% CI 5.1 to 8) per 100 000 per year for children under 5 years of age. The incidence of cases classified by ICD-10 code search was 16.0 (95% CI 11.3 to 21.8) per 100 000 per year for infants and 4.0 (95% CI 2.9 to 5.3) per 100 000 per year for children under 5 years. Year-by-year differences between the CPT and ICD-10 code search incidences (figures 1 and 2) were not statistically significant (all p>0.05, data not given).

Figure 1.

Incidence* of non-fatal AHT in infants admitted to Starship hospital from within the metropolitan Auckland and Northland region. Cases determined by the CPT compared with cases classified by ICD-10 code search. *Annual incidence (dots) and line of best fit (Poisson regression). Grey: AHT as determined by the CPT; black: AHT as classified by the ICD-10 code search. AHT, abusive head trauma; CPT, child protection team; ICD-10, International Classification of Diseases, Tenth Revision.

Figure 2.

Incidence* of non-fatal AHT in children under 5 years of age admitted to Starship hospital from within the metropolitan Auckland and Northland region. Cases determined by the CPT compared with cases classified by ICD-10 code search. *Annual incidence (dots) and line of best fit (Poisson regression). Grey: AHT as determined by the CPT; black: AHT as classified by the ICD-10 code search. AHT, abusive head trauma; CPT, child protection team; ICD-10, International Classification of Diseases, Tenth Revision.

The trend line of non-fatal AHT incidence in infancy, as determined by the CPT, represented an annual change of −4.9% (95% CI −12.6 to +3.5%) (figure 1). As classified by the ICD-10 code search the trend line represented an annual change of −8.7% (95% CI −18.3 to +2.2%). These trends are not significantly different from each other or from no change. For every year, except 2010, the annual incidence as determined by the CPT was within the 95% CI of the ICD-10 code search incidence regression line (data not shown). A similar pattern of incidence trend lines was present in children under 5 years (figure 2).

Discussion

The principal finding of our study was the relatively low sensitivity of the broad definition of non-fatal AHT, identifying only 67% of cases of non-fatal AHT (95% CI 57 to 75). This contrasts with a study from a tertiary paediatric hospital in Pittsburgh,²⁰ where the sensitivity was 92% (95% CI 85.8 to 96.2) and specificity 96% (95% CI 92.3 to 99.7). One possible explanation for the disparity may be that explicit diagnostic codes such as ‘child abuse unspecified’ (ICD-9: 995.5, ICD-10: T74.9), ‘child physical abuse’ (ICD-9: 995.54, ICD-10: T74.1), ‘shaken baby syndrome’ (ICD-9: 995.55—not available in Australasia) and ‘other child abuse and neglect’ (ICD-9: 995.59, ICD-10: T74.8) were applied more consistently in Pittsburgh, where clinical records are electronic. Electronic systems may help ensure that diagnoses of abuse are recorded clearly and transferred rapidly to discharge coders. However, most New Zealand hospitals have neither electronic clinical records nor a specialist CPT. The reduced sensitivity of the broad definition in our study may, therefore, provide a more accurate representation of front-line clinical reality. Conversely, the low number of false positives suggests that this definition is unlikely to overestimate the incidence of non-fatal AHT. It could be used in New Zealand to monitor the incidence of AHT over time, provided it is recognised that it will produce a material underestimate of the true incidence.

A limitation of our study is that it was undertaken at a single tertiary children’s hospital with a specialist CPT, where both clinical and coding practice may differ from other hospitals in New Zealand. This could limit the generalisability of our findings to other districts. This is particularly relevant as Starship admits a significant proportion of AHT in New Zealand with the rest divided among a number of hospitals providing care to much smaller populations, where abuse codes are likely to be used much less frequently. If anything, this would be expected to reduce the sensitivity of the broad definition even further.

Most other studies using the broad definition have not tested it against a gold standard. One other study, from Alaska, reported a sensitivity of 91% (95% CI 0.82 to 1.0) and a specificity of 99% (95% CI 0.98 to 1.0) when compared with a gold standard which involved identifying cases from multiple data sources (including police, court, child protection and social media), which were then reviewed by child protection specialists.³³ In contrast, the low sensitivity in our study is consistent with other literature on the limitations of ICD codes for tracking child maltreatment. A recent study of ICD-10-CM codes compared against the gold standard of assessment by a CPT at a paediatric level I trauma centre in Texas, found that for inpatients with all forms of child maltreatment the sensitivity was 55.6% (95% CI 41.4 to 69.1) and the specificity 78.6% (95% CI 69.0% to 91.7%).³⁴

In the USA, the incidence of AHT calculated using the broad definition has been relatively consistent in different studies. The first, from the CDC, found an annual incidence of 32.3 (95% CI 27.0 to 37.6) per 100 000 infants and 8.6 (95% CI 7.2 to 10.0) per 100 000 for children under 5 years of age.¹³ Shanahan et al found a national annual incidence of 38.8 (95% CI 36.6 to 41.1) per 100 000 infants.¹⁵ Dias et al described a maximum incidence in infants in Pennsylvania of 46.1 (95% CI 42.1 to 50.4) per 100 000.¹⁸ In contrast, Fujiwara et al described an incidence in Canada of 15.5 (95% CI 13.6 to 17.6) per 100 000 infants,¹⁹ very similar to the incidence obtained in our study using the same ICD code set.

The best estimate of the incidence of AHT in New Zealand came from a national prospective study from 2000 to 2002, where the ‘maximum’ estimate was 31.1 (95% CI 23.1 to 40.9) per 100 000 infants under 1 year of age. That study included fatal AHT (mortality rate of 13.5%),²¹ but as discussed in detail elsewhere, probably underestimated the incidence by as much as 20%.³ Nonetheless, that figure is consistent with the incidence determined in this study from the assessment of the CPT, and further supports a conclusion that the use of the broad definition in New Zealand will underestimate the true incidence of AHT.

The exclusions from the broad definition (birth trauma, sexual abuse, coagulation defects, unintentional gun-related injury and congenital malformations) may be unwarranted. We found three cases of definite AHT classed as false negatives by these exclusions—two with secondary coagulopathies caused by AHT and one with a congenital malformation of the brain as well as AHT. Although the exclusions exist to ensure that rare or alternative presentations are not miscoded as AHT, no such error occurred in 10 years of admissions to New Zealand’s sole tertiary paediatric centre. We suggest that the broad definition could be modified for the purposes of public health surveillance by dispensing with these exclusions.

We also found a significant difference between false and true negatives in the proportion with code X59 (exposure to unspecified factor). Many false negatives had no history of trauma, which from a clinical perspective is often suspicious for abuse.²⁶ It may be that X59 is often used to cover circumstances which should raise concerns about abuse, although it is important to note that it also appears in true negatives. The performance of the definition in New Zealand might be improved by training clinicians to document conclusions about child abuse more clearly and expeditiously, while seeking further clarification about coding practice, particularly regarding use of the X59 code in a setting of suspected abuse.

Finally, it is important to note a major limitation of using population surveillance data to track the incidence of a rare condition: it is difficult to judge the efficacy of a prevention programme. Random clustering of cases can have a large effect on a trend line without representing a genuine change in incidence.¹⁵ The trend lines in figures 1 and 2 should not be over-interpreted. The most that can be said is that they may provide some grounds for optimism (in contrast to at least one earlier study)²⁶ that the rate of AHT in New Zealand is not continuing to increase. Any use of the broad definition for population surveillance will need to take a cautious view of year-to-year fluctuations.

Data availability statement

Data are available upon reasonable request. Anonymised data may be shared upon request from the corresponding author (ORCID 0000-0002-8813-8877). A data sharing agreement will be generated, and all data will be anonymised prior to sharing.

Ethics statements

Patient consent for publication

Not applicable.

Ethics approval

Ethics approval was received from the New Zealand Health and Disability Ethics Committee (project ID 4135) and the Te Toka Tumai Auckland Institutional Research Office.