The Trauma Audit Filters Trial (TAFT)

Survey of the field

Trauma, defined as the clinical entity composed of physical injury and the body's associated response, kills almost five million people each year. This is more than the total number of annual deaths from malaria, tuberculosis, HIV/AIDS, and maternal conditions combined. More than 90% of trauma deaths occur in low and middle income countries (LMIC), and about 30-50% of these deaths have been reported to occur in hospital. Research show that almost 11% of the global burden of disease estimated using disability adjusted life years are due to trauma, and that disability adjusted life years from road traffic trauma has increased by 35% in the last 25 years.

The United Nation now vows to reduce the number of road traffic deaths by 50% by 2020. Although primary prevention will play a major role in achieving this, several international actors, including the World Health Organization (WHO), emphasise the importance of strengthened trauma care. A considerable body of research on the strengthening of trauma care attributes improvements in trauma patient outcomes to the implementation of quality improvement programmes, defined as programmes to improve "health care through monitoring the process of care and measuring outcomes".

For example, a recent single centre study from Australia showed a reduction in in-hospital mortality from 16 to 10% after the implementation of a quality improvement programme. This programme included interventions such as a protocol for trauma team activation, massive transfusion protocols, case reviews and the recording and follow up audit filters. Similarly, research from Thailand and Pakistan show reduced mortality after the implementation of such programmes. Unfortunately the heterogeneity of interventions included in most quality improvement programmes makes it hard to draw conclusions about the impact of individual components.

Despite this heterogeneity, audit filters constitute a common denominator across different quality improvement programmes. Such filters can be defined as "pre-identified variables that are routinely tracked to identify whether accepted standards of care are being met". Hence, audit filters are also referred to as quality indicators or key performance indicators. The concept of audit filters in trauma care originates from the American College of Surgeons guidelines on trauma care . They defined 22 filters and the basic idea was that each filter should represent a "sentinel event" associated with poor patient outcome.

In theory, audit filters may catalyse improvements in care and therefore the use of such filters is now widespread in trauma care organisations. Research on the impact of implementing audit filters on patient outcomes is however scarce. One widely cited study from Thailand demonstrated a reduction in preventable mortality after the revision of audit filters; however, this study was not deemed to be of high enough quality to merit inclusion in the only Cochrane review conducted on the subject. In fact, the authors of that review were not able to identify a single study of high enough quality.

The push to include audit filters in efforts to strengthen trauma care is strong, despite the remarkable lack of evidence that such filters improve patient outcomes. The WHO even includes audit filters as a core technique in their guidelines for trauma quality improvement programmes . Considering the often extremely limited resources in the contexts of their intended audience, primarily decision and policy makers in LMIC, it is crucial that the recommendations of potentially costly interventions are evidence based. Therefore, the research question is, does institutional implementation of audit filters reduce mortality in adult trauma patients?

Hypothesis

The implementation of audit filters will lead to: 1) identification of potentially correctable deficiencies; 2) subsequent correction of identified deficiencies; and 3) ultimate improved clinical care and reduced mortality.

Study design

A controlled interrupted time series trial. This design is generally considered to be the strongest quasi experimental design when a randomised controlled trial is not feasible. In brief, the trial will be composed of three phases. First there will be an observation phase spanning one year, during which outcome baselines will be established. Then there will be an implementation phase during which locally relevant and appropriate audit filters will be developed and implemented and an audit filter review board will be elected. This phase will be six months. Finally, there will be an intervention phase for two years when the audit filters will be reviewed at monthly meetings.

Setting

This trial will be conducted in four university hospitals in India. There are several reasons why India will be used as a model. First, India accounts for 20-25% of all trauma mortality globally and efforts to strengthen trauma care in the country are urgently needed. The results of this trial will thus be highly relevant to its participants as well as the community at the large. Second, none of the centres that will participate use audit filters today. Hence, the time of implementation will be precisely controlled and thereby one source of bias will be substantially reduced. All four centres are public university hospitals with demonstrated research capacity. Each centre has approximately 1500 beds and all clinical specialities relevant for trauma care available in house. The trial will start in 2017 and continue for four years after ethical approval from participating centres.

Source and method of participant selection

One project officer at each centre will be employed to enrol participants and collect data. The project officer will be posted in the emergency department and enrol consecutive patients that fit the eligibility criteria. She or he will work day, evening, and night duties according to a rotating random schedule so that all possible shifts are covered during the course of each month. Each shift will be eight hours, out of which the project officer will spend six in the emergency department and two conducting follow up of admitted and discharged patients. The project officer will work five shifts per week. The aim of this model for participant selection and data collection is to enrol a representative sample of each centre's population of eligible participants, maximising data quality by having the project officers record outcomes and covariates, while at the same time minimising data collection costs and potential observer bias.

Study setup

As alluded to previously, the trial will have three distinct phases:

observation, implementation, and intervention.

Observation phase The observation phase will be used to establish a baseline for primary outcome and secondary outcome. The data collection procedures outlined in section "Source and method of participant selection" will be applied in all four centres.

Implementation phase (Month 12-18)

Two centres will randomly be chosen as control centres and the remaining two will be chosen as intervention centres. Data collection continues as during the observation phase in both groups.

Control centres: No change compared to the observation phase.

Intervention centres: The goal of the implementation phase in the intervention centres will be to develop and implement locally relevant and appropriate audit filers, as well as a system to identify and implement solutions to identified deficiencies. To achieve this goal a participatory and multidisciplinary approach will be used. First a meeting and two day course will be hosted at each intervention centre to present the background and rationale for using audit filters to improve trauma care. Representation by all disciplines and specialities involved in trauma care, including clinical specialities, nursing, rehabilitation and administration will be strived for.

The two day course on trauma quality improvement processes (TQIP) including audit filters will include the following objectives: 1) Gaining an understanding of TQIP and a familiarity with the evidence supporting it; 2) Gaining an understanding of implementation options for different clinical contexts; 3) Gaining an understanding of the application of audit filters in TQIP; 4) Training on the implementation of targeted corrective action plans; 5) Explanation of the utility of ongoing data collection to evaluate interventions ("closing the loop"). This course will provide an overview of techniques used for hospital-based TQIP, focusing on the concepts of audit filter application, and processes of flagged event review.

After the course participants on professor, associate professor, and assistant professor level, i.e. 10-15 people, will be invited to take part in an anonymous online Delphi survey. The Delphi approach means that the survey will be delivered in multiple rounds, and that participants receive feedback on the previous round's results before a new round is initiated. In each round the participants will be provided with a set of audit filters and will be asked to rate each filter on a scale from 1 to 10 (where 1 represent useless and 10 very useful), provide written comments on suggested filters, and finally suggest new audit filters, and provide a written rationale for each.

In the first round participants will be provided with a set of audit filters used in other contexts. All audit filters that get a median rank of ≥7 as well as all filters provided by participants will be included in subsequent rounds. The Delphi survey will be terminated when either consensus or stability of group responses is reached. Consensus is defined as the situation when all included audit filters have a median rank of ≥9. Stability is defined as the situation when no significant change in audit filter ranks occurs between two rounds. If stability is reached then all audit filters with a median rank of ≥9, or the five audit filters with the highest ranks, will be implemented.

The results of the Delphi survey will be presented in each intervention centre at a second meeting. This first post-survey meeting will only host the survey participants. After this meeting each centre will be asked to elect an audit filter review board. This board will review audit filter data, identify potentially correctable deficiencies and solutions. The board will also have the mandate to implement potential these solutions. Once an audit filter review board has been elected it will inform the relevant staff cadres of the audit filter implementation and review mechanisms.

The audit filters will then be implemented. Data on audit filters will be collected and collated by a second project officer, employed by us, in each of the intervention centres. Project management will participate in the three first audit filter review meetings. At these meetings each flag will be assessed by reviewing the patient's record. The plan, do, study, act (PDSA) cycle typically employed to implement and evaluate interventions to improve processes of care will be taught.

Intervention phase (Month 18-42)

Control centres: No change compared to the observation and implementation phase.

Intervention centres: The audit filter review board will conduct monthly review meetings. Project management will not participate in these meetings. The audit filter review board will be free to add or remove audit filters and implement and evaluate interventions as they choose. The meetings will be documented in standardised protocols.

Covariates

Throughout the trial data on the following covariates will be collected to allow describing participants as well as adjust analyses for potential differences in case mix between centres and over time:

• Age in years. Recorded by project officers.
• Sex, coded as male or female. Recorded by project officers.
• Systolic blood pressure. Recorded by project officers using dedicated equipment.
• Diastolic blood pressure. Recorded by project officers using dedicated equipment.
• Heart rate. Recorded by project officers using dedicated equipment.
• Blood oxygen saturation. Recorded by project officers using dedicated equipment.
• Respiratory rate. Recorded by project officers by manual counting during one minute.
• Level of consciousness using the Glasgow coma scale and Alert, Voice, Pain or Unresponsive scale (AVPU). Recorded by project officers.
• Mechanism of injury according to chapter XX of ICD-10. Recorded by project officers.
• Injuries sustained, extracted from patient records by project officers to enable anatomical injury severity scoring using the injury severity score (ISS).
• Transferred from another health facility. Recorded by project officer by asking participant or participant's relative.
• Mode of transport. Recorded by project officer by asking participant or participant's relative.
• Date and time of injury, to calculate delay between injury and presentation to participating centre. Recorded by project officer by asking participant or participant's relative.
• Date and time of computed tomography investigation. Recorded by project officer from the participant's hospital record or by direct observation.
• Date of ultrasonography investigation. Recorded by project officer from the participant's hospital record or by direct observation.
• Date when mechanical ventilation was started and stopped. Recorded by project officer from the participant's hospital record or by direct observation.
• Date, time and type of surgery, defined as any surgical intervention performed under general or regional anesthesia. Recorded by project officers from the participant's hospital record or by direct observation.

Statistical methods and analyses

A segmented general additive model (GAM) will be used to assess the impact of institutional implementation of audit filters on outcomes. The GAM approach will allow accommodation of potential nonlinearity in intervention effect while adjusting for autocorrelation, seasonality and case mix differences. Data from intervention and control centres will be pooled separately. Each observation will represent averaged data for all patients enrolled during one month instead of data for an individual patient.

A second analysis will be conducted using patient level data. In both analyses the primary interest is detecting a change in trend between the observation and intervention phases, as well as between control and intervention centres, for each of the outcomes. Furthermore, estimation of intervention effect on both mortality and quality of life will be possible by calculating the difference between predicted outcomes assuming no intervention effect and the observed outcomes with intervention. R will be used for statistical analyses, adopting a 95% confidence level and 5% significance level. The Holm procedure will be used to adjust for multiple tests.

Subgroup analyses

• Patients with major trauma, defined as any patient who is admitted for more than three days with an ISS of >15, or who dies within three days of admission, or who dies between arrival and admission.
• Patients with life-threatening, but potentially salvageable trauma, defined as any patient with 15<ISS<24.

Sensitivity analyses

• Exclusion of the implementation phase from the analysis model.
• Data will be pooled for the observation and intervention phase and analysed using a pre-post design.