Table 1

Health services research techniques with applications to trauma and acute care surgery research

Hierarchical modeling
Application	Approach	Interpretation
Measuring hospital variation	Output ‘cluster’-level effects from hierarchical model for the outcome Graphical result is a caterpillar plot	Quantifies and depicts overall variation between hospitals not explained by differences in patient characteristics.
Median OR	V_A=hospital-level variance	Median increase in odds of outcome if patient was treated and randomly selected hospital of higher risk.
Proportional change in variance	V₁=hospital-level variance in model containing patient-level variables only V₂=hospital-level variance in model with factor of interest added	Proportion of hospital variation in outcome due to the factor of interest.
Intraclass correlation coefficient	V_A=hospital-level variance V_I=individual-level (patient-level) variance	Proportion of all variation in outcome that is attributable to differences between hospitals.
Advanced confounder control
Application	Approach	Interpretation
Propensity score (PS)	PS reflects probability of treatment Derived for each patient using logistic regression model adjusted for all important factors available Matching or weighting provides risk-adjusted association between treatment and outcome	Must be done carefully. Important to report methodology and balance between matched groups in keeping with best practices. Limitations and potential for unmeasured confounding must be discussed.
Instrumental variable (IV)	IV is highly correlated with treatment but unrelated to outcome Good IV is typically unavailable, therefore surrogate ‘area-level’ measure of the process under evaluation is typically derived	Association between IV and outcome should approximate causal relationship. Validity of the IV must be demonstrated. Limitations of IV used must be discussed.
Geospatial analysis
Application	Approach	Interpretation
Access-to-care	Straight-line distance or time (eg, air transport) Network analysis to derive time or distance along public roads (eg, road transport) Service areas represent areas within defined distance or time categories	Estimates distance or time for injured patients to reach hospital via ground or air transport. Can be used to quantify % of populations with specific categories of access-to-care.
Hot spot analysis	Outcomes or events are aggregated within the geographic unit of measurement (eg, ZIP codes) Hot spot analysis compares value of each geographic area with those of surrounding geographic areas Hot or cold spots are identified as 90%, 95%, or 99% outlier areas	Hot spots or cold spots should be evaluated to understand what is contributing to significantly higher-than-expected or lower-than-expected outcome rates. Implications are proposed for changes in policy or trauma system design.