Article Text

Download PDFPDF

Injury surveillance in accident and emergency departments: to sample or not to sample?
  1. A Morrison,
  2. D H Stone
  1. Paediatric Epidemiology and Community Health (PEACH) Unit, Department of Child Health, University of Glasgow, Royal Hospital for Sick Children, Yorkhill, Glasgow G3 8SJ
  1. Correspondence to: Anita Morrison.


Objectives—To establish whether injury surveillance based on sampling strategies is as valid as total patient surveillance.

Methods—Canadian Hospitals Injury Reporting and Prevention Program (CHIRPP) data for 1996 were retrospectively analysed using five sampling frames. Proportions for key variables were calculated for each sample, then compared with the proportions for the total population of patients.

Results—Two of the five sampling frames produced statistically significant differences from the total population, which can be explained by seasonal variations. However, no significant differences were observed between the remaining three samples and the total population.

Conclusions—A well planned and executed sampling strategy can generate as valid data as total patient surveillance, obviating the need for data collection on every patient presenting with an injury or poisoning. In practice, however, systematic sampling can be difficult to implement and sustain, counterbalancing the economic advantages.

  • surveillance
  • sampling

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Evidence suggests that injury surveillance in accident and emergency departments is a valuable and achievable objective.1–,3 Such systems provide a proactive mechanism for routinely monitoring injury incidence, identifying risk factors, stimulating preventive programmes, and evaluating their effectiveness.3 However, the implementation and operation of such a system involves additional, often scarce, resources. In addition to the costs of system installation and maintenance, a considerable investment in personnel is usually required to collect, computerise, and analyse information.1, 2

Although hospitals may be sampled to provide a representative data set across regions or nations, the majority of injury surveillance systems collect information on all patients presenting to the designated accident and emergency departments,1, 4–,7 making these systems relatively expensive. This paper examines whether it is necessary to include all patients in injury surveillance, or whether a sample offers a valid, and cheaper, alternative. Data collected during 1996 by the Canadian Hospitals Injury Reporting and Prevention Program (CHIRPP) system at the Royal Hospital for Sick Children in Glasgow were analysed retrospectively to examine the extent to which the use of various sampling frames would have resulted in representative samples, as reflected by the frequency distribution of selected key variables.


CHIRPP is a computerised information system that collects data on all patients presenting with injuries or poisoning to accident and emergency departments. This system has been in operation at the Accident and Emergency Department, Royal Hospital for Sick Children, Yorkhill in Glasgow since 1993.1 During the calendar year 1996, 7940 CHIRPP forms were completed (over 90% of eligible patients). Several sampling frames were applied to these data. The sampling frames were largely arbitrary, although they had been identified as logistically possible: sample 1: every 10th attendance (n=794); sample 2: every eighth day (n=1032); sample 3: once a week (week day) (n=1104); sample 4: once a week (weekend) (n=1123); and sample 5: four months out of 12 (n=2245).

Proportions for the following key variables were calculated: age, sex, injury type, injury location, activity at the time of injury, and the body part injured. For each sampling frame, the proportions for each sample were compared with the proportions for the total population of injured children recorded on CHIRPP. Confidence intervals were calculated, and differences between proportions examined using the common proportion and the standard error.7 The null hypothesis assumed no differences in the proportions between the samples.8


Three statistically significant differences were found between the proportions generated by the sampling frames and the events observed in the total population of injured children (table 1). Significantly more children sustained an injury while playing and significantly fewer sustained injuries in the school playground in the four month sample (sample 5) than in the total population. The third difference was observed between the weekend sample (sample 4) and the total population: the proportion occurring in the school playground was significantly higher in the total population. No significant differences were found between the other sampling frames (samples 1, 2, and 3) and the total population.

Table 1

CHIRPP data 1996: sampling frames compared; values are per cent (95% confidence interval)


The significant differences observed in the proportions between two of the systematic sampling frames and the total population of injured children may be explained by seasonal variation. July (included in sample 5) encompasses the school holidays, when it is likely that there will be an increase in the number of injuries sustained while playing in and around the home and a decrease in the number of injuries sustained in the school playground. This is probably also true for the weekend sample (sample 4) with respect to injuries sustained in the school playground.

No other significant differences were found, suggesting that a well planned and executed sampling strategy, such as sampling every 10th attendance or collecting data every eighth day, can generate data of equal quality to surveillance including all patients. The use of a systematic sampling strategy may relieve the pressure on both personnel and other resources at accident and emergency departments by obviating the need for data collection on every patient presenting with an injury or poisoning. Such sampling strategies would not affect the overall quality of information available on a medical record because the CHIRPP data collection sheet is completed separately from routine record keeping.

Systematic sampling is also economically attractive. Adopting a sampling strategy for the CHIRPP system in Glasgow would result in cost reductions of approximately 30%. The costs of installing CHIRPP, staff training, and data analysis would remain the same. However, considerable savings would be made in the data collection, coding, and computerisation phases. Such a reduction in the costs of running an injury surveillance system may entice other public health organisations and agencies to investigate the possibilities of introducing such a system in their locality.

Nevertheless, in practice, systematic sampling may present logistical problems that counterbalance the economic attractions. Staff may forget to include every patient, or more seriously, select cases in a biased manner (according to perceived severity, for example). Other disadvantages include the inability of a sample to provide a comprehensive profile of injuries, especially unusual or densely clustered events, and the unsuitability of a sample based surveillance system for clinical purposes, such as audit or medicolegal review.

Of course, surveillance based on data collected from just one hospital (whether sampled or not) is a relatively poor alternative to a population based system. However, if as this study suggests, a systematic sampling strategy is not detrimental to the quality of data collected at any one hospital, the adoption of a systematic sampling strategy in a representative sample of hospitals in a locality may result in better quality epidemiological information than a single hospital collecting information on all injury events.

This paper has demonstrated that a systematic sampling strategy can generate as valid data on the pattern of injuries presenting to accident and emergency departments as total population surveillance. However, there are both advantages and disadvantages to adopting sampling strategies. Each department must assess whether sampling is appropriate depending on their available resources and information requirements.


We would like to thank Mr Doraiswamy, consultant at the Accident and Emergency Department, Royal Hospital for Sick Children (RHSC), Yorkhill and the department staff for their assistance in implementing and operating the CHIRPP system. Thanks also go to the clinical audit team at RHSC for inputting and managing the data.