seasonality of child homicide

5
Seasonality of Child Homicide Antoinette L. Laskey, MD, MPH, Jonathan D. Thackeray, MD, Sophia R. Grant, MD, and Patricia G. Schnitzer, PhD Objective To determine whether there is seasonal variation (by season and month of year) in homicides among young children. Study design Homicide deaths in children #5 years (n = 797), as identified in death certificates from 5 states (In- diana, Ohio, Missouri, Oklahoma, and Washington), were analyzed for a seasonality effect in the years 1999–2006. Season and month of calendar year were used as categorical variables. A c 2 goodness of fit test was used to com- pare observed and expected proportions of homicides for the entire population. Stratified analyses of children dying before age 2 years and those dying between 2 and 5 years of age were also conducted. Results The occurrence of homicide was proportional to the time interval in each category (P = .05). For the total population, there was no statistically significant variation between the expected and observed percentage of homi- cides by either season or month of calendar year (P = .46 and P = .74, respectively). For the stratified analyses, there was no statistically significant variation between expected and observed percentages of homicides by either sea- son or month of calendar year for either population. Conclusion There is no seasonality to child homicides by month of year or season of year among young children in the examined population. (J Pediatr 2010;157:144-7). I nflicted injuries are the leading cause of traumatic death in children younger than age 4 years, with approximately 1500 children dying at the hands of caregivers each year. 1 Seventy-six percent of all child abuse–related deaths occur in children younger than 4 years of age, and 42% occur among infants younger than 1 year of age. Anecdotally, many people believe that child abuse increases around the winter holidays or during the winter months, particularly severe cases of abuse such as abusive head trauma (AHT), which is usually inflicted by shaking or blunt impact. 2 For this reason, some authors have referred to the winter months as ‘‘shaken baby season’’ 3 or ‘‘infanticide season.’’ 4 It has been suggested that this increase in child abuse is re- lated to increased caregiver stress at this time of year. Although there has been some research into the existence of a seasonal effect on AHT, evidence sufficient to support this commonly-held belief has not been well demonstrated, in large part because of very small sample sizes. 3,5 In contrast to the inconclusive findings of studies specific to children, significant seasonal effects in adult homicides, suicides, and violent crime rates have been well-documented. 6-10 For example, an Oklahoma study on injury-related death with medical examiner data demonstrated a summer-time peak in unintentional injury and homicide deaths, 6 and other studies have shown a seasonal effect on trauma injuries in general. 11-14 These findings in adult populations suggest that seasonality is at least con- ceivable in pediatric populations. Previous research has failed to demonstrate that there is a ‘‘shaken baby season.’’ The purpose of this study was to examine whether there is seasonal variation in homicides within a large population of young children so that we might develop more effective prevention measures. Methods A convenience sample of 5 states (Indiana, Ohio, Oklahoma, Missouri, and Washington) provided death certificate data for all children younger than age 5 years who died between January 1, 1999, and December 31, 2006, with homicide listed as the man- ner of death. The overwhelming majority of child homicides are due to child abuse (ie, committed by a caregiver). 15 Data col- lected from death certificates included age of child at time of death, exact date of death, manner of death (all were homicide), sex, race, and ethnicity. We analyzed our data for a seasonality effect with both season and month of year used as cat- egorical variables. Seasons were calculated in the following way: winter included December 1–February 29 (total number of days = 722); spring included March 1–May 31 (total number of days = 736); summer included June 1–August 31 From the Children’s Health Services Research, Department of Pediatrics, Indiana University School of Medicine, Indianapolis, IN (A.L.), Nationwide Children’s Hospital, The Ohio State University College of Medicine, Columbus, OH (J.T.), Department of Pediatrics, University of Oklahoma College of Medicine, Tulsa, OK (S.G.), Sinclair School of Nursing, University of Missouri, Columbia, MO (P.S.) The views expressed within this article are those of the authors and do not necessarily represent the views of Indiana University. The authors declare no conflicts of interest. 0022-3476/$ - see front matter. Copyright Ó 2010 Mosby Inc. All rights reserved. 10.1016/j.jpeds.2010.01.013 AHT Abusive head trauma FBI SHR Federal Bureau of Investigation Supplementary Homicide Reports NVSS National Vital Statistics System 144

Upload: antoinette-l-laskey

Post on 02-Sep-2016

216 views

Category:

Documents


0 download

TRANSCRIPT

Seasonality of Child Homicide

Antoinette L. Laskey, MD, MPH, Jonathan D. Thackeray, MD, Sophia R. Grant, MD, and Patricia G. Schnitzer, PhD

Objective To determine whether there is seasonal variation (by season and month of year) in homicides amongyoung children.Study design Homicide deaths in children #5 years (n = 797), as identified in death certificates from 5 states (In-diana, Ohio, Missouri, Oklahoma, and Washington), were analyzed for a seasonality effect in the years 1999–2006.Season and month of calendar year were used as categorical variables. A c2 goodness of fit test was used to com-pare observed and expected proportions of homicides for the entire population. Stratified analyses of children dyingbefore age 2 years and those dying between 2 and 5 years of age were also conducted.Results The occurrence of homicide was proportional to the time interval in each category (P = .05). For the totalpopulation, there was no statistically significant variation between the expected and observed percentage of homi-cides by either season or month of calendar year (P = .46 and P = .74, respectively). For the stratified analyses, therewas no statistically significant variation between expected and observed percentages of homicides by either sea-son or month of calendar year for either population.Conclusion There is no seasonality to child homicides by month of year or season of year among young childrenin the examined population. (J Pediatr 2010;157:144-7).

Inflicted injuries are the leading cause of traumatic death in children younger than age 4 years, with approximately 1500children dying at the hands of caregivers each year.1 Seventy-six percent of all child abuse–related deaths occur in childrenyounger than 4 years of age, and 42% occur among infants younger than 1 year of age. Anecdotally, many people believe that

child abuse increases around the winter holidays or during the winter months, particularly severe cases of abuse such as abusivehead trauma (AHT), which is usually inflicted by shaking or blunt impact.2 For this reason, some authors have referred to thewinter months as ‘‘shaken baby season’’3 or ‘‘infanticide season.’’4 It has been suggested that this increase in child abuse is re-lated to increased caregiver stress at this time of year. Although there has been some research into the existence of a seasonaleffect on AHT, evidence sufficient to support this commonly-held belief has not been well demonstrated, in large part becauseof very small sample sizes.3,5

In contrast to the inconclusive findings of studies specific to children, significant seasonal effects in adult homicides, suicides,and violent crime rates have been well-documented.6-10 For example, an Oklahoma study on injury-related death with medicalexaminer data demonstrated a summer-time peak in unintentional injury and homicide deaths,6 and other studies have showna seasonal effect on trauma injuries in general.11-14 These findings in adult populations suggest that seasonality is at least con-ceivable in pediatric populations.

Previous research has failed to demonstrate that there is a ‘‘shaken baby season.’’ The purpose of this study was to examinewhether there is seasonal variation in homicides within a large population of young children so that we might develop moreeffective prevention measures.

AHT Abusive head trauma

FBI SHR Federal Bureau of Investi

NVSS National Vital Statistics S

144

Methods

A convenience sample of 5 states (Indiana, Ohio, Oklahoma, Missouri, and Washington) provided death certificate data for allchildren younger than age 5 years who died between January 1, 1999, and December 31, 2006, with homicide listed as the man-ner of death. The overwhelming majority of child homicides are due to child abuse (ie, committed by a caregiver).15 Data col-

From the Children’s Health Services Research,Department of Pediatrics, Indiana University School ofMedicine, Indianapolis, IN (A.L.), Nationwide Children’sHospital, The Ohio State University College of Medicine,Columbus, OH (J.T.), Department of Pediatrics,University of Oklahoma College of Medicine, Tulsa, OK

lected from death certificates included age of child at time of death, exact date ofdeath, manner of death (all were homicide), sex, race, and ethnicity. We analyzedour data for a seasonality effect with both season and month of year used as cat-egorical variables. Seasons were calculated in the following way: winter includedDecember 1–February 29 (total number of days = 722); spring included March1–May 31 (total number of days = 736); summer included June 1–August 31

(S.G.), Sinclair School of Nursing, University of Missouri,Columbia, MO (P.S.)

The views expressed within this article are those of theauthors and do not necessarily represent the views ofIndiana University. The authors declare no conflicts ofinterest.

0022-3476/$ - see front matter. Copyright � 2010 Mosby Inc.

All rights reserved. 10.1016/j.jpeds.2010.01.013

gation Supplementary Homicide Reports

ystem

Table I. Characteristics of sample (n = 797)

Frequency Percent*

Male 402 56%Age at death

<12 months 357 45%12 m-<24 m 169 21%24 m-<36 m 112 14%36 m-<48 m 71 9%48 m-<60 m 60 8%60 m-<72 m 26 3%

RaceWhite 405 51%African-American 250 33%Hispanic 20 3%Asian 8 1%Native American 14 2%Multi-Racial 7 1%Unknown/Unreported 83 10%

Year of Death1999 89 11%2000 90 11%2001 110 14%2002 107 13%2003 105 13%2004 100 13%2005 92 12%2006 104 13%

*Not all add to100% because of rounding.

Table II. Expected versus observed homicides by season

Season

Observedhomicides,

wholepopulation

(%)*

Observedhomicides,<2 yearsold (%)†

Observedhomicides,$2 yearsold (%)z

Expectedhomicides (%)

Winter 26.73 25.48 29.37 24.71Spring 23.21 24.52 20.82 25.19Summer 25.35 23.76 27.88 25.19Fall 24.72 26.24 21.93 24.92

*c2, P = .46.†c2, P = .11 (observed vs expected).zc2, P = .80 (observed vs expected).

Vol. 157, No. 1 � July 2010

(total number of days = 736); and fall included September1–November 30 (total number of days = 728).

The null hypothesis for our study was that the occurrenceof homicide is uniform throughout the year, such that thenumber of homicides should be proportional to the time in-terval (or number of days) in each category (season ormonth). Expected numbers of homicides were calculated asdescribed in Tiihonen et al.10 In brief, the expected percent-ages were calculated by determining the exact number of daysthat fell within a given time period (month, accounting forleap years; or season, accounting for leap years) and dividingby the number of categories (12 for months and 4 for sea-sons). These percentages were then applied to the total num-ber of homicides to arrive at the expected numbers in eachcategory. A c2 goodness of fit test was used to compare ob-served and expected proportions of homicides for the entirepopulation. If the ratio of actual homicides to expected ho-micides was 1, there was no variation in that interval (seasonor month). If 1 did not fall within the 95% confidence inter-val, the null hypothesis was rejected because there was a statis-tically significant variation in that interval. Because it hasbeen previously demonstrated that a majority of abusivehead trauma deaths occur in the under-2-years age group,stratified analyses of children dying before the age of 2 years(n = 526) and those dying between 2 and 5 years of age (n =269) were also conducted. This study was considered exemptby the institutional review board of Indiana University.

Results

There were 797 deaths classified as homicide among childrenin participating states during the study period. Of these chil-

dren, 402 (56%) were male, and 526 (66%) were youngerthan 2 years old (Table I). Homicide deaths occurreduniformly throughout the year, regardless of month orseason (Tables II and III). The percent of homicides byseason and month of year in comparison with thecalculated expected percent are shown in Figures 1 and 2(available at www.jpeds.com). In our analysis of the totalpopulation, there was no significant difference in theoccurrence of homicide in either season or month ofcalendar year (c2, P = .46 and P = .74, respectively). Therewas also no significant difference in the occurrence ofhomicide by season or month of calendar year for eitherthe children younger than 2 years old (c 2, P = .11 andP = .41, respectively) or the 2- to 5-year-olds (c2, P = .80and P = .34, respectively) (Tables II and III).

Discussion

Our study indicates that, in a large sample of children underthe age of 5 years, homicide does not vary by season ormonth, indicating that children are likely to be equally atrisk of homicide death throughout the year. This lack of var-iance suggests that the stressors triggering physical childabuse resulting in homicide might occur at any time, regard-less of month or season. Prevention efforts should thereforefocus on addressing caregiver responses to common stressorssuch as crying, toileting accidents, and normal childhood be-havioral issues such as temper tantrums. Anticipatory guid-ance and education on the dangers of stress and frustrationshould be directed to all adults who provide care for youngchildren, throughout the year.

Using identical methods, Tiihonen et al10 identified sea-sonal variation in adult homicides in Finland, with peaks insummer months and troughs in winter months. Other stud-ies with a variety of statistical methods used to analyze sea-sonal patterns have also identified summer peaks in adulthomicides with a variety of hypotheses about why this wouldbe.6,8,9,16,17 Conversely, other studies with a variety of statis-tical methods and a broad mix of data sources have failed tofind seasonal variation in adult homicides.7,18,19

Some authors and healthcare providers have suggestedthat abuse and homicide escalate during the winter monthsand associated holidays. McCleary and Chew,4 using

145

Table III. Expected versus observed homicides bymonth

Month

ObservedHomicides,

wholepopulation (%)*

ObservedHomicides,<2 yearsold (%)†

ObservedHomicides,$2 yearsold (%)z

ExpectedHomicides (%)

January 8.78 7.79 10.78 8.49February 7.65 5.89 11.15 7.73March 7.65 7.98 7.06 8.49April 8.03 8.56 7.06 8.21May 7.53 7.98 6.69 8.49June 8.16 7.41 9.29 8.21July 9.79 9.13 10.78 8.49August 7.40 7.22 7.81 8.49September 8.03 8.37 7.43 8.21October 8.91 9.70 7.43 8.49November 7.78 8.17 7.06 8.21December 10.29 11.79 7.43 8.49

*c2, P = .74.†c2, P = .41 (observed vs expected).zc2, P = .34 (observed vs expected).

THE JOURNAL OF PEDIATRICS � www.jpeds.com Vol. 157, No. 1

a different analytical strategy, found that there was an in-creased risk of child homicides during winter months for in-fants, toddlers, and preschoolers (OR 1.25, 1.26 and 1.16,respectively). It may be that an effect of this size cannot bedemonstrated without a sample size comparable with theone used in McCleary’s study (ie,�14 000). Given the annualnumber of child homicides though, this effect may not be rel-evant on an annual basis. Another explanation for the differ-ence in results could be the different datasets used. Researchhas shown that homicide data from the FBI SHR can varyfrom homicide data collected in the National Vital StatisticsSystem (NVSS), which gathers data from death certificates.FBI SHR data can be prone to underreporting, and NVSSdata can be prone to misclassification.9 It is possible thatthe reason child abuse is believed to increase during holidayperiods is that providers associate these cases with a memora-ble point in time. Abused children seen on nondescript daysof the year may be memorable, but those that present ona memorable day (eg, Christmas, New Year’s) may be easierto recall because of association with the holidays.

This study is the first of its size to use death certificate datato explore seasonality of child homicide. As in all studies,there are some limitations. One is that death certificates donot reliably capture all child homicides.20,21 It is unlikely,however, that there would be a systematic difference in homi-cide cases being correctly coded as such by date of death.Missed homicide cases on death certificates are therefore un-likely to influence the outcome of these analyses. A secondlimitation is that we selected all pediatric homicides withoutdiscriminating homicide by caregiver versus non-caregiver.Although most pediatric homicides in the age range studiedare the result of abuse by a caregiver, our numbers likely rep-resent a small overestimate of pediatric homicides in the con-text of child maltreatment. Another possible limitation is thatthese findings may not be generalizable to other states, al-though the diverse geographic and demographic scope ofour participating states should counter this. Finally, it is pos-

146

sible that the length of time examined may not be sufficient toaccurately capture trends. Our large sample size, however,should counter this possible confounder, and it is importantto note that a longer study period may have allowed secularchanges to effect observed patterns of abuse and homicide.

Our analysis of nearly 800 homicides of children under theage of 5 years from 5 states failed to demonstrate any seasonalor monthly variation. Although it has been suggested thatthere is a ‘‘shaken baby season,’’ a seasonal trend was notseen in our data. Although national data demonstrate thatAHT is the leading cause of child abuse–related deathsamong children younger than 4 years of age, multiple studieshave demonstrated the limitations of using death certificatesto identify child abuse–related deaths because these causes ofdeath are invariably underreported in death certificatedata.15,20,21 Furthermore, there is not a specific cause of deathcode for AHT deaths. These limitations prevented us fromdetermining the actual number of AHT deaths from amongall child homicides in this dataset. It is possible that the subsetof AHT deaths do increase around the holidays, but a largerstudy with multiple data sources or a data source known to bemore precise, such as child fatality review program data,would be necessary to address this question. Our data suggestthe need for year-round prevention efforts with regard tochild abuse–related deaths from all causes. n

We wish to thank Dr. Brian Johnston, Tim Stump, Kelly Haberkorn,the Indiana State Department of Health, and Amy Lewis Gilbert.

Submitted for publication Oct 9, 2009; last revision received Dec 2, 2009;

accepted Jan 7, 2010.

References

1. Child Maltreatment 2007. In: U.S. Department of Health and Human

Services; 2007[cited 2009 July 8]; Available from: http://www.acf.hhs.

gov/programs/cb/pubs/cm07/figtab4.htm.

2. Christian CW, Block R. Committee on Child Abuse and Neglect. Abu-

sive head trauma in infants and children. Pediatrics 2009;123:1409-11.

3. Daly SE, Connor SM. Seasonal variations in the incidence of suspected

shaken baby syndrome. Int J Trauma Nurs 2001;7:124-8.

4. Mccleary R, Chew KSY. Winter is the infanticide season: seasonal risk for

child homicide. Homicide Studies 2002;6:228-39.

5. Barlow KM, Minns RA. Annual incidence of shaken impact syndrome in

young children. Lancet 2000;356:1571-2.

6. Goodman RA, Herndon JL, Istre GR, Jordan FB, Kelaghan J. Fatal in-

juries in Oklahoma: descriptive epidemiology using Medical Examiner

data. South Med J 1989;82:1128-34.

7. Maes M, Cosyns P, Meltzer HY, De Meyer F, Peeters D. Seasonality in

violent suicide but not in nonviolent suicide or homicide. Am J Psychiatr

1993;150:1380-5.

8. Maes M, Meltzer HY, Suy E, De Meyer F. Seasonality in severity of

depression: relationships to suicide and homicide occurrence. Acta

Psychiatr Scand 1993;88:156-61.

9. Hakko H, Rasanen P, Tiihonen J. Increasing homicide rate in Finland ac-

companied by decreasing seasonality over the period 1957-95. Soc Sci

Med 1998;47:1695-8.

10. Tiihonen J, Rasanen P, Hakko H. Seasonal variation in the occurrence of

homicide in Finland. Am J Psychiatr 1997;154:1711-4.

11. Loder RT, O’Donnell PW, Feinberg JR. Epidemiology and mechanisms

of femur fractures in children. J Pediatr Orthop 2006;26:561-6.

Laskey et al

July 2010 ORIGINAL ARTICLES

12. Rainey S, Cruse CW, Smith JS, Smith KR, Jones D, Cobb S. The occur-

rence and seasonal variation of accelerant-related burn injuries in Cen-

tral Florida. J Burn Care Res 2007;28:675-80.

13. Rising WR, O’Daniel JA, Roberts CS. Correlating weather and trauma

admissions at a level I trauma center. J Trauma 2006;60:1096-100.

14. Puljula J, Savola O, Tuomivaara V, Pribula J, Hillbom M. Weekday dis-

tribution of head traumas in patients admitted to the emergency depart-

ment of a city hospital: effects of age, gender and drinking pattern.

Alcohol Alcohol 2007;42:474-9.

15. Herman-Giddens ME, Brown G, Verbiest S, Carlson PJ, Hooten EG,

Howell E, et al. Underascertainment of child abuse mortality in the

United States. JAMA 1999;282:463-7.

16. Ambade VN, GodboleHV, Kukde HG. Suicidaland homicidaldeaths: a com-

parative and circumstantial approach. J Forensic Legal Med 2007;14:253-60.

Seasonality of Child Homicide

17. Lester D. Temporal variation in suicide and homicide. Am J Epidemiol

1979;109:517-20.

18. Abel EL, Welte JW. Temporal variation in violent death in Erie County,

New York, 1973–1983. Am J Forensic Med Pathol 1987;8:107-11.

19. Abel EL, Strasburger EL, Zeidenberg P. Seasonal, monthly, and day-

of-week trends in homicide as affected by alcohol and race. Alcohol

Clin Exp Res 1985;9:281-3.

20. Ewigman B, Kivlahan C, Land G. The Missouri child fatality study:

underreporting of maltreatment fatalities among children younger

than five years of age, 1983 through 1986. Pediatrics 1993;91:

330-7.

21. Crume TL, DiGuiseppi C, Byers T, Sirotnak AP, Garrett CJ. Underascer-

tainment of child maltreatment fatalities by death certificates, 1990–

1998. Pediatrics 2002;110:e18.

147

Figure 1. Observed versus expected percent homicides byseason.

Figure 2. Observed versus expected percent homicides bymonth.

THE JOURNAL OF PEDIATRICS � www.jpeds.com Vol. 157, No. 1

147.e1 Laskey et al