PHE 505 Module 8 Milestone Four Example

How the Study Would Work: Methods and Analysis Plan for Linking Helmet Use to Head Trauma in Scooter Crashes

[Student Name]

Southern New Hampshire University

PHE 505: Research Methods in Public Health

Final Project Milestone Four

[Instructor Name]

[Date]

The organization, setting and figures below are a composite written as a model document. No real employer, client, colleague or patient is described.

How the Study Would Work: Methods and Analysis Plan for Linking Helmet Use to Head Trauma in Scooter Crashes

Design

The study looks back through emergency department charts, analyzed as a case-control comparison within injured riders. Cases are riders with any head injury; controls are riders injured elsewhere on the body. The question is whether helmet nonuse is more common among cases than controls, after adjustment. Many studies in the bicycle helmet meta-analysis by Olivier and Creighton (2017) used this logic, which suits settings where only injured people can be studied. Its key assumption is that helmets affect the chance of a head injury but not the chance of other injuries, so riders with other injuries reflect helmet use among those who crash.

Setting and Participants

The setting is the city's two emergency departments, which share an electronic record system. Participants are patients aged 18 or older whose visit followed a crash while they were on a rented electric scooter. Patients injured as pedestrians struck by scooters, riders of privately owned scooters and patients transferred in from other hospitals will be excluded. Hulley et al. (2013) advise writing inclusion and exclusion criteria so that a second reviewer would select the same records; each criterion above has a written definition and an example.

Finding the Records

Records will be identified in two ways. First, a search of diagnosis codes for injuries involving standing micro-mobility devices. Second, a keyword search of triage notes for scooter, the local operators' brand names and e-scooter. Because coding of scooter injuries was inconsistent when the fleets first launched, the keyword search is expected to find records the codes miss. Every candidate record will be reviewed to confirm eligibility.

Data Collection

Two trained abstractors will extract the variables defined in Milestone Three using a standard form with written rules. Both will independently abstract a random 10% of records, and agreement on helmet use and head injury will be measured with kappa; a value below 0.7 would trigger retraining and revision of the rules. Abstractors will record whether helmet status was documented at all, so missingness can be analyzed. Abstracted data will live in a protected database inside the hospital network, stripped of identifiers.

Before the full review, a pilot of 50 records will test the form, the rules and how often helmet status is missing. If more than half the pilot records lack helmet information, the team will consider adding a short observation study of riders as a second source of evidence.

Sample Size and Power

The preliminary search found about 480 scooter injuries in three years, roughly 430 in adults. In published studies, about 4% to 6% of injured riders wore helmets and about 40% had head injuries. Using G*Power (Faul et al., 2007), a study with 430 riders, of whom about 26 wore helmets, would have only about 50% power to detect an odds ratio of 0.5, the size reported for bicycle helmets. Reaching 80% power requires about 850 riders, including roughly 50 who wore helmets.

Two steps would close the gap. The review will add the university's urgent care clinic, which treats many minor scooter injuries, and extend the review period to four years. Together these are expected to yield about 850 adult riders. If they fall short, the study will report its estimate with its confidence interval and describe the result as imprecise, not as evidence of no effect.

Table 1. Power under different sample sizes

Note. Assumes 6% helmet use, 39% head injury among unhelmeted riders and a two-sided alpha of 0.05.

Analysis Plan

Descriptive statistics will summarize riders' characteristics by helmet use and by head injury. The main analysis will use multivariable logistic regression with head injury as the outcome and helmet use as the exposure, adjusting for age, sex, alcohol involvement, nighttime arrival and collision with a vehicle, and will report adjusted odds ratios with 95% confidence intervals. The same model with traumatic brain injury as the outcome will test the secondary hypothesis, and adding a product term for helmet status by night arrival will explore the third. Model fit will be checked, and correlated covariates will be examined before inclusion. Analyses will be run in R.

Sensitivity Analyses

Three sensitivity analyses will test how fragile the result is. Records with missing helmet status will be assigned first to the helmeted group and then to the unhelmeted group. The main model will be repeated excluding riders who collided with vehicles, whose injuries may overwhelm any helmet effect. And it will be repeated with only the emergency department records, to see whether adding urgent care patients changed the estimate.

Threats to Validity

Selection bias is possible, since lightly hurt riders who skipped medical care never enter the data; the urgent care clinic reduces but does not remove this. Information bias may arise if clinicians record helmets more often when they see a head injury. Unmeasured confounding, especially speed and riding experience, cannot be adjusted for. And results from one university city may not generalize. The study will be reported using the STROBE checklist so that readers can judge these threats (von Elm et al., 2007).

Ethics and Data Protection

The protocol will request a waiver of individual consent and of authorization under federal privacy rules, on the grounds set out in the ethics journal: minimal risk, impracticability of contacting former patients and strong safeguards. Only the hospital analyst will see identifiers, which are removed before the abstraction file is created. Each patient's dates will move by a random offset, and results will be reported only in groups of ten or more. The data will be destroyed three years after publication, following the hospital's retention policy.

Timeline

Review board approval is expected in months one and two, abstraction rules and the pilot of 50 records in month three, full abstraction in months four to seven, analysis in month eight and the report in month nine. Creswell and Creswell (2018) recommend building time for problems into research plans, so one month of slack is included before the report.

References

Creswell, J. W., & Creswell, J. D. (2018). Research design: Qualitative, quantitative, and mixed methods approaches (5th ed.). SAGE.

Faul, F., Erdfelder, E., Lang, A.-G., & Buchner, A. (2007). G*Power 3: A flexible statistical power analysis program for the social, behavioral, and biomedical sciences. Behavior Research Methods, 39(2), 175-191. https://doi.org/10.3758/BF03193146

Hulley, S. B., Cummings, S. R., Browner, W. S., Grady, D. G., & Newman, T. B. (2013). Designing clinical research (4th ed.). Lippincott Williams & Wilkins.

Olivier, J., & Creighton, P. (2017). Bicycle injuries and helmet use: A systematic review and meta-analysis. International Journal of Epidemiology, 46(1), 278-292. https://doi.org/10.1093/ije/dyw153

von Elm, E., Altman, D. G., Egger, M., Pocock, S. J., Gøtzsche, P. C., & Vandenbroucke, J. P. (2007). The Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement: Guidelines for reporting observational studies. The Lancet, 370(9596), 1453-1457. https://doi.org/10.1016/S0140-6736(07)61602-X