Adverse incidents in a paediatric ICU: a model to identify latent risk factors

Introduction

Successful risk management requires a robust system for adverse incident (AI) reporting and analysis. Although many ICU studies have focused on AI categorisation and examination of factors immediately preceding the AI, few have investigated the potential contribution from latent or hidden factors (e.g. poor staff supervision). This phenomenon is well known in other complex, nonmedical systems such as the aviation and nuclear industries.

Aim

To establish latent risk factors for AI occurrence in a tertiary PICU.

Methods

Data were collected prospectively from 730 consecutive, 12-hour nursing shifts over a 12-month period (816 patient episodes, standardised mortality ratio 0.77). Incidents were reported and categorised using a standardised format. Factors potentially contributing to the occurrence of an AI were classified according to temporal aspects (time of day, weekend, etc.), bed occupancy, patient dependency, patient flux, nursing and medical skill-mix, junior medical staff supervision, nursing agency (locum) use and a weighted, composite score quantifying several factors that may compromise the clinical supervisory role of the nurse in charge (e.g. patient death on a shift, logistical issues with support staff such as porters, coverage of meal breaks, etc.). Two logistic regression models were constructed: one examining the relationship between these variables and the type of AI; the second examined potential interaction effects between clinically related variables (e.g. 'patient workload' combines bed occupancy with patient dependency).

Results

Two hundred and eighty-four AI occurred during 30% (220/730) of shifts, of which 181 were patient related and 103 unit or staff related. Patient-related AI were categorised as: drug errors (55), intravenous/arterial line problems (37), equipment issues (32), patient injury (26), standard of care (21) and self-extubation (10). One hundred and thirty-four (74%) of the patient-related CIs resulted in actual harm to the patient, of which 49 (27%) were deemed serious.

Both unit-related and patient-related AI were more common during the day shift (the period of greatest ICU activity). Unit-related AI were also associated with factors compromising the senior nurses' supervising role, OR 1.31 (95% CI 1.03–1.68), while patient-related AI were less common with increased junior doctor supervision, OR 0.61 (95% CI 0.40–0.91). Factors associated with the various categories of patient AI (drug errors, self-extubation, etc.) included: patient dependency, bed occupancy, number of admissions/discharges per shift, increased nursing agency use, and absence of a senior sister on the shift. The second regression model, examining interaction effects, demonstrated an interaction between nursing supervision factors for unit-related AI, and patient workload factors for patient-related AI. Both models demonstrated excellent goodness of fit (Hosmer Lemeshow P > 0.10).

Conclusion

AI are common and are associated with many latent factors including time of day, nursing and medical supervision, and patient workload. This model may provide a focus for strategies aimed at reduction in AI.