Objective To reduce the number of routine chest radiographs (CXRs) done in a tertiary care intensive care unit (ICU).
Methods Using a quality improvement approach, we measured the number of CXRs done per patient-day before (15 June 2010–15 June 2011) and after (15 June 2011–15 June 2012) a multipronged intervention in a 15-bed medical–surgical ICU in a 350-bed tertiary care teaching hospital. We studied a total of 1492 patients who were admitted to this ICU—738 patients during the preintervention period and 754 patients during the postintervention period. Interventions were education for the ICU house staff, developing indications for routine CXRs on the computer order-entry system, and visual posters/signage to remind ICU staff that there were no indications for routine, daily CXRs. The primary outcome was the number of CXRs per patient-day, but we also measured CTs of the chest, mechanical ventilator days, length of ICU stay and ICU and hospital mortality.
Results There were 0.73 CXRs per patient-day done during the preintervention period and 0.54 CXRs per patient-day done during the postintervention period, a 26% reduction. There were no differences between the periods in age, sex or severity of illness (Acute Physiology and Chronic Health Evaluation (APACHE) II score) of the patients, number of chest CTs, mechanical ventilator days, length of ICU stay and ICU or hospital mortality.
Conclusions A quality improvement that includes education, reminders of appropriate indications and computerised decision support can decrease the number of routine CXRs in an ICU.
- Critical care
- Healthcare quality improvement
- Hospital medicine
- Health services research
Statistics from Altmetric.com
In the intensive care unit (ICU), many tests and investigations are ordered and done routinely. For example, until recently, the American College of Radiology (ACR) had recommended routine daily chest radiographs (CXRs) for all mechanically ventilated patients, and the use of further CXRs if necessary.1 Although a few studies support selective use of CXRs, most observational studies published before 2000 support the use of daily CXRs because they have found that these radiographs affect daily management.2–9 For example, in a study by Henschke et al,5 up to 65% of routine CXRs may have led to differences in management or treatment decisions.
Recent clinical studies have investigated whether this practice is still warranted, especially because the practice of critical care has evolved, including changes in mechanical ventilation strategies that favour lower tidal volumes and improved training in imaging modalities such as ultrasonography. A recent cluster-randomised trial in 21 French ICUs showed that an ‘on-demand’ strategy (vs daily routine strategy) was associated with a 32% reduction in the number of CXRs without any effect on other diagnostic or therapeutic interventions.10 In the same study, there was no difference in duration of mechanical ventilation or stay in ICU, or mortality. A meta-analysis by Oba and Zaza11 demonstrated that elimination of daily routine CXRs did not affect either hospital (OR, 1.02; 95% CI 0.89 to 1.17) or ICU (OR 0.92, 95% CI 0.76 to 1.11) mortality. Based on these findings and those of other recent studies,10–17 the ACR recently changed their recommendations,18 stating that ‘routine daily chest radiographs are not indicated for patients with acute cardiopulmonary problems’. Despite the new evidence and changes in recommendations, many physicians, including physicians at our own institution, still continue to order daily routine CXRs in ICU patients.
Considering the increasing interest in promoting efficiency in the provision of healthcare and stopping overinvestigation and healthcare waste,19–22 we decided to use a quality improvement approach to reduce the number of routine CXRs done in our ICU. Using a similar on-demand strategy, we aimed to reduce the number of CXRs per patient-day by 25% within 1 year.
The project team for this work was composed of the authors of this paper. Before initiating this quality improvement initiative, we obtained approval from the research ethics board to conduct this study. This initiative was done in a 15-bed ICU, at a 350-bed tertiary teaching hospital. Our patients included a mixture of medical and surgical patients but no trauma, thoracic surgical, neurosurgical or lung/liver transplant patients. We decided to focus our efforts at one institution to test the specific changes, and to improve the probability of implementing changes. We identified potential targets for change by mapping out the processes leading to CXRs being done (figure 1). Most CXRs in the ICU are ordered by house staff (residents and medical students), who are on a 1- or 2-month rotation in ICU during their clinical training. Laboratory tests and radiographic investigations at our institution are ordered through Sunrise Clinical Manager (SCM V.4.5, Eclipsys), a computer-based system for patient care. We directed our interventions primarily at education of house staff, and made changes to the computerised order-entry system to encourage reduction in the ordering of routine CXRs. During this study, there were no other new interventions to reduce use of nasogastric tubes, to prevent ventilator-associated pneumonia, to change scheduling of medical staff or to change use of ultrasound for diagnosis.
This was a before–after design. We implemented our interventions prospectively, and identified 15 June 2011 as the date that our changes would be initiated. We collected data during the preintervention period (15 June 2010–14 June 2011) and the postintervention period (15 June 2011–15 June 2012). We included all patients who were admitted to our ICU during this time period.
Quality improvement interventions
Before beginning our interventions, we identified the major stakeholders in this process and collaborated with the ICU attending physicians, radiologists, information technology (IT) department, nurses, respiratory therapists and ICU house staff. The study investigators met with the ICU attending physicians group to develop a set of acceptable indications for ordering a CXR in the ICU, using the ACR recommendations, as well as indications described in other previous studies.10 ,18 ,23 These indications were: unexplained new cardiopulmonary symptoms or signs; suspected new pneumonia; suspected pneumothorax; suspected new pleural effusion; insertion of endotracheal tubes, feeding tubes, chest tubes or central venous catheters and suspected malposition or malfunction of existing tubes. We also created an ‘other’ category to allow physicians and house staff to specify indications or unique circumstances that may not have been included in the list above. ‘Routine’ or ‘mechanical ventilation’ were not considered valid indications for ordering a CXR. We also met with the local IT department to incorporate this list of indications (and ‘non-indications’) into our computerised order-entry system.
Our interventions to change practice (figure 1) were: education of house staff (medical students, residents, fellows) at the beginning of each month regarding current evidence about the value of routine CXRs, a prompt in the ICU computerised order-entry system to allow only acceptable indications when ordering CXRs (figure 2) and the display of posters in the ICU which reinforced the acceptable indications. These interventions have been incorporated into regular ICU operations and are still in effect.
Methods of evaluation
Using the ICU database at our institution, we collected patient demographics, procedures in the ICU, duration of mechanical ventilation, length of ICU stay and ICU and hospital mortality for each of the patients admitted during both periods. Using the hospital radiology database, we collected the number of routine priority and urgent/stat CXRs and the number of chest CTs done during the same periods.
By dividing the total number of CXRs done by the total number of patient-days in the ICU for each period, we calculated the total number of CXRs done per patient-day, which was the primary measure of interest. The 95% confidence interval (CI) for these estimates was obtained by treating patient-days as a fixed variable and bootstrapping on the number of tests done per day. We generated 95% CIs from a sample of 1000 mean number of tests per patient-day values; the confidence limits reported are the 2.5 and the 97.5 percentile. We also subdivided this variable by the priority of the test—routine, urgent or ‘stat’ CXRs. Balancing measures were the number of chest computerised tomograms (CTs) done per patient-day and the number of new catheters or tubes (central venous catheters, nasogastric/gastric feeding tubes, intubations and chest tubes) inserted. We analysed data over a 1-year period in both groups to minimise the effects of seasonal variations of disease (eg, influenza, pneumonias). Due to delays in obtaining data, no data feedback was provided during the postintervention period.
Statistical analysis was done using SAS V.9.4 software. To compare numbers of CXR and other events per patient-day between the two periods, we calculated the ratio of the number of events per patient-day during the postintervention period divided by the number of events per patient-day during the preintervention period and the associated CI. A ratio that has a CI that does not include 1 is statistically significant. To illustrate the trends, data were plotted using values of tests per patient-day on a weekly basis and a U-chart was created for CXRs per patient-day. Rules for special cause variation24 were: any points outside the control limits, a run consisting of seven or more consecutive points on either side of the centre line, a trend consisting of seven or more consecutive points moving upwards or downwards, four out of five consecutive points that hug the centre line and a cyclical pattern.
There were 1492 patients admitted to the ICU between 15 June 2010 and 15 June 2012; 738 patients were admitted before the interventions and 754 patients afterwards. There were no differences in demographic characteristics (age, sex) or severity of illness (Acute Physiology and Chronic Health Evaluation (APACHE) II scores) between the preintervention and postintervention patient groups (table 1).
CXRs per patient-day
A total of 4066 CXRs were done before the interventions, corresponding to 0.73 CXRs/patient-day. After the intervention, there were 2980 CXRs done, corresponding to 0.54 CXRs/patient-day (tables 2 and 3).
The ratio of post/pre for overall number of CXRs/patient-day was 0.74 (95% CI 0.71 to 0.76), which indicates a 26% reduction. The largest reduction was in the number of routine CXRs done (post/pre ratio of 0.33, 95% CI 0.25 to 0.40). There was no significant increase in the number of urgent CXRs done (post/pre ratio of 1.03, 95% CI 0.91 to 1.15) but there was a modest increase in the number of stat CXRs done (post/pre ratio of 1.18, 95% CI 1.05 to 1.30); this corresponds to a very small number of additional stat CXRs (table 2). There was an overall shift from the baseline in the number of CXRs/patient-day done after implementation of our quality improvement initiatives, and only three points of special cause variation (both above and below the mean; figure 3). We were unable to compare indications for CXR between the two periods because documentation of indications was not required during the preintervention period.
There was no change in the number of chest CT scans done during the entire period of observation, and there was also no change in the number of catheters or tubes inserted (figure 4). Furthermore, there were no changes in days of mechanical ventilation, ICU length of stay, ICU mortality or hospital mortality between the two periods (table 1).
At our institution, the average cost of a CXR in the ICU is $C25, including radiologist fees for interpreting the films. Therefore, we estimate that we saved about $C27 150 per year as a result of this intervention.
After a multipronged quality improvement intervention, we achieved a substantial reduction in total CXRs/patient-day. Our interventions were associated with a reduction in the number of CXRs for at least 1 year after they were implemented. We surpassed our aim of reducing the total number of CXRs/patient-day in the ICU by achieving a reduction of 26%, and a 67% reduction in routine priority CXRs. There were no changes in the number of chest CT scans, or in any of the patient outcomes that we measured. Our study was done in one Canadian ICU that has a mixture of intubated and non-intubated medical and surgical patients. The findings from this study can be generalised to many ICUs because our interventions were simple and because many centres already use computer-based order entry.
Other studies, including a more recent meta-analysis published after this study was completed, have reported similar reductions in CXRs/patient-day without compromising ICU length of stay or mortality.10 ,11 ,25 Krinsley17 implemented a form to record the reasons for ordering CXRs; this intervention led to a 22% reduction in CXRs. However, CXRs could still be ordered without completion of the form in their study. The advantages of our study included use of a force-function in the computer order-entry system to ensure compliance, and to remind and educate the ordering physician directly at the time that the test is ordered. We also combined this intervention with other educational strategies. One advantage of this approach is that physicians are prompted to evaluate the need for tests on a daily basis instead of defaulting to ordering all tests daily; the result is a culture change in test ordering. Another strength of our study is that it was conducted by six medical residents; they accomplished an improvement in care, and also learned about quality improvement by practising it.
Reducing CXRs in the ICU may have economic and potentially other benefits. Although we had a modest savings in costs, if the cost of CXRs is greater, there may be even greater savings. In addition, portable lung ultrasonography has been shown to detect pneumothorax much more reliably than portable CXRs.26 Funds saved from reducing the number of CXRs in the ICU could be diverted to other initiatives, such as the purchase of a portable ultrasound machine or other ICU equipment, which may improve patient care or safety in other ways.
The interventions implemented in this study were neither costly nor labour-intensive, making this on-demand strategy easily applicable in other similar ICUs. In addition to reduction in costs, avoiding unnecessary CXRs could also reduce exposure to radiation, reduce risk of adverse events associated with repositioning patients for CXRs, improve efficiency of daily ICU patient rounds and reduce nursing/aide time to repositioning patients.
There are a few limitations to our study. Our interventions were implemented almost simultaneously making it difficult to determine the individual impact of each change. A strategy in which we introduced each initiative individually over time would have allowed us to evaluate this quality improvement intervention in multiple plan-do-study-act cycles to implement only the most beneficial interventions. However, we felt that a multipronged approach would improve the probability of a change in practice. Some of the intervention strategies that we used may be more difficult to sustain in the long-term. Educational initiatives are difficult to sustain as they require a dedicated person to ensure that this message is passed on monthly. Second, we used only one computerised order-entry system—it is possible that the changes we made might be more difficult in other computerised systems. Third, we did not collect data on whether the diagnostic yield of the CXRs done increased, or whether the CXRs done led to any changes in care, but this had already been done in previous studies.10 Finally, we did not tabulate the indications for ordering CXRs. However, the list of indications differed considerably between the two periods, so comparing these indications would have been problematic.
Our study has demonstrated the feasibility for individual centres to implement this kind of a quality improvement project in the ICU. For example, computer-based approaches can be powerful when used strategically. There are still many other tests which may be ordered routinely or unnecessarily, and therefore, a similar strategy could be implemented for these tests. Recently, the Critical Care Societies Collaborative, in collaboration with the Choosing Wisely campaign, has advocated for the appropriate ordering of diagnostic tests in response to a clinical question.22
Simple interventions, such as education and computer order entry, are effective in reducing the number of CXRs ordered without any adverse impact on patient outcomes or other processes of care.
The study authors would like to thank the attending ICU physicians, nurses and radiographers for their assistance.
Contributors Each of the authors contributed to the design of the study and to the review of the manuscript for important intellectual content. ES, ML, MQ, YK and SS were also responsible for the implementation of improvements involved in this study and together they drafted the first version of the manuscript. MN and HW were also responsible for acquisition of data from clinical databases and for all of the statistical analysis. All authors approved the final version of the manuscript and are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.
Competing interests None declared.
Ethics approval UBC Providence Health Care Research Ethics Board.
Provenance and peer review Not commissioned; externally peer reviewed.
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.