Chapter 2. ST-PRA Development

Proactive Risk Assessment of Surgical Site Infection in Ambulatory Surgery Centers

To achieve project objectives, the AIR team designed a study that leveraged information gleaned through four unique and important quantitative and qualitative data sources:

  • Known risk factors and their rate estimates from the extant literature on SSIs.
  • National databases of inpatient care, ambulatory surgical care, and emergency department care to provide procedure volumes, institutional characteristics, patient demographics, and complication estimates.
  • Site visits to four local ASCs with different organizational arrangements, to provide contextual elements for the sociotechnical component of the risk modeling.
  • Technical experts' input to further enhance the models and provide additional expertise on the sociotechnical elements of the risk models.

In this chapter, we begin with an overview of ST-PRA followed by the methods involved in collecting information from each data source. The final section of this chapter focuses on the methodology employed for building the fault tree model, how probability estimates from the literature and other sources were utilized in the model, and the resulting fault tree model.

Overview of Sociotechnical Probabilistic Risk Assessment

Probabilistic Risk Assessment (PRA) is an engineering tool that was developed in the 1970s to quantify risks and identify threats to the safety of nuclear power plants.1 Subsequently, it has been applied in a variety of settings, which range from aerospace to manufacturing to natural disasters.2 PRA is a systematic methodology that proactively identifies the major risk points in a system. It utilizes both quantitative and qualitative data to "map" the risks associated with adverse outcomes.1-2

PRA is a hybrid between qualitative process analysis techniques and quantitative decision-support models.1-3 PRA involves a detailed "deductive" process analysis method that utilizes logical relationships and probability theory to construct a model (a "fault tree") of how the various risk points interact with one another and either individually or collectively combine to contribute to the overall adverse outcome. PRA has several major strengths, because it:4

  • Represents a broad perspective and includes contextual elements, such as operating procedures, system factors, and human factors, in the risk model.
  • Is proactive, identifying the possible adverse events before they actually occur, thus enabling the decision maker to make targeted interventions for preventing those events.
  • Allows, through the use of logical relationships and Bayesian probabilities, the modeling of complex interactions and dependencies among the multiple risk points that may lead to the adverse outcome.
  • Allows the uncertainty associated with error rate estimates to be incorporated into the model through sensitivity analysis.
  • Allows an assessment of risk and a prioritization of risk reduction interventions based on sequences that have the highest probability of occurrence, thus providing a roadmap of targeted interventions.
  • Is dynamic, in that PRA can incorporate new estimates of probability based on uncertainty using Bayes' theorem.

ST-PRA expands the basic PRA model by accounting for human performance.3-4 Most work involves the interactions of people, systems, and technology, and ST-PRA accounts for each of these elements. The challenge in this approach is determining the probabilities associated with human breakdowns that contribute to adverse outcomes.

The process mapped by a ST-PRA model incorporates factors that are internal to the process and factors that are external to the process.4-5 For example, in the setting of infection prevention in the operating room, internal process factors include choosing the incorrect disinfectant or not using appropriate skin disinfection procedures. Factors external to the process include the ability of policies and procedures to direct best practices, the safety norms for following such policies, and the ability of team members to give and understand crucial communications. Consequently, ST-PRA can disentangle the impact of factors that are related to individuals from the impact of factors that are related to institutions or to the system. In this way, ST-PRA addresses what has previously been described as a major limitation of isolated database analyses in which the interactions of different-level processes occur simultaneously. To ensure that ST-PRA captures all possible process factors, it is important to scour several sources of data with the aim of building a process map. In the next section, we describe the sources of data scanned for this purpose.

Return to Contents

Data Sources

Exhibit 1 depicts the data types (i.e., quantitative or qualitative) and the sources used in the development of the ST-PRA fault tree model for this study. Each data source informed the data collection effort for the other sources in an iterative fashion. That is, information gleaned during the literature review informed ways to analyze the databases; information collected during the site visits or during the TEP meeting informed additional data analyses and literature searches.

Literature Review

The AIR team first conducted an extensive literature review of peer-reviewed and grey literature regarding the potential risk factors associated with the development of an SSI in surgical procedures generally and for arthroscopy specifically. As noted in an earlier, unpublished interim report for this project, arthroscopy of the knee was identified as the surgical procedure that would serve as the focal point for this project. In addition to helping establish the risks and finalize the inputs to the ST-PRA models, the literature also provided discrete probability estimates and ranges for inclusion in the models and for the sensitivity testing, which is the final step in model development. In this section, we describe the process for searching and abstracting both peer-reviewed and grey literature.

Peer-reviewed literature. The peer-reviewed literature search was limited to the literature published in English and, initially, only that published since 2000. The date limit was amended to include literature published as early as 1985, because it quickly became clear that important work related to the estimates of risk dictating current clinical practices dates as far back as the late 1980s. Similarly, the search was expanded to include research conducted beyond the United States, to include important work related to the risks of SSI conducted in Europe, the United Kingdom, and Australia.

 

Search engines used to conduct the literature review included PubMed®, the Cochrane Collaborative, Cumulative Index to Nursing and Allied Health Literature (CINAHL), and other search engines, as appropriate. As with any literature review, we began by using a series of keyword search terms within the categories of interest to assist us in better understanding the extant literature on this topic. These search terms, by category, are presented in Exhibit 2.

Because there are no standardized search terms for the analysis of SSIs in the ambulatory surgery setting, we used additional categories and search terms as the literature review proceeded or as specific risk points or estimates needed to be identified. The intent of this work was not to create an exhaustive literature review around each term, but to ensure that we were incorporating relevant work into the risk models.

 

After identifying potential articles in the literature, we reviewed the abstracts to determine their relevance for inclusion. We reviewed the entire article for those elements that had relevance. For example, often an entire article would be reviewed only to identify a single probability estimate related to a specific content area (e.g., risk of SSIs in morbidly obese patients). We also reviewed the reference lists in articles, to ensure that we were being as inclusive as possible. We established general inclusion and exclusion criteria to help establish when an article would or would not be included in our literature review. Examples of these inclusion and exclusion criteria are provided in Exhibit 3.

Grey literature. To enhance the review of extant peer-reviewed literature, we included a review of grey literature to provide important information for inclusion in the risk models. This category of literature includes Web-based presentations, articles, and white papers. For this part of the literature search, we used Google™, Google Scholar™, and Bing Internet search engines. We also surveyed the project team and our technical experts to identify additional sources of information not contained within the peer-reviewed literature.

As a final step in this part of the review, we conducted a targeted search of Web sites known for their independent contribution to improvement efforts or standards of care, to see whether they had contributed any information on general or ambulatory surgical risks. Examples of the organizations included in this part of the review are:

  • Agency for Healthcare Research and Quality.
  • Institute for Healthcare Improvement.
  • American College of Surgeons.
  • Ambulatory Surgery Center Association.
  • Robert Wood Johnson Foundation.
  • Association of Perioperative Registered Nurses.
  • Centers for Medicare and Medicaid Services.
  • The Joint Commission on Accreditation of Healthcare Organizations.

We entered all relevant literature into a database to assist with creation of a bibliography and to support the specific risk estimates in model building. The resulting literature review is presented in Appendix A.

Database Analysis

The next source of process factor information was AHRQ's national databases for surgery and inpatient admissions. Discharge databases play an important role in studying SSIs in the ASC environment. Although ASCs provide an important setting for the performance of ambulatory procedures, they are constrained in that complications, particularly infections that occur in these settings, may receive followup care in the physician's office, in the emergency department, or, for severe cases, in the hospital. Hence, ASC databases themselves may provide only limited information on complication or infection rates, because the treatment for these complications and infections often occurs outside the ASC setting. To assist with modeling the risks associated with SSIs in the ambulatory surgery setting, it was necessary to analyze important, publicly available extant databases to see whether relevant probability estimates for complications originating in the ASC environment could be obtained.

First, we sought to gather quantitative information from the different types of locations that might care for complications originating in the outpatient surgery environment (i.e., private physician offices, ASCs, emergency departments, and hospitals).

Private Physician Offices. There are currently no publicly available datasets for analysis from physician's offices. Thus we were unable to identify patients who received surgery in an ASC, experienced a complication, and returned to their private surgeon/physician for followup on the complication. This is a major limitation to the analysis of surgical complications, including SSIs, stemming from surgeries conducted in the ambulatory setting.

Ambulatory Surgery Centers. AHRQ's Healthcare Cost and Utilization Project (HCUP) State Ambulatory Surgery Databases (SASD) were analyzed to identify the most common surgeries, the demographic profile of individuals receiving them, and the institutional profile of locations in which they are performed. Specifically, the 2006-2008 SASDs for Maryland, New Jersey, and California were used in this study. Because the SASDs for California had data on both hospital-based and free-standing ASCs, and included many variables that were of interest to this study, we analyzed the 2008 California SASD extensively.* We then extended this analysis to years 2006 and 2007, which produced results similar to those for 2008. This analysis resulted in a list of the top 10 procedures for hospital-based and free-standing ASCs. We then analyzed the distribution of these procedures for hospital-based ASCs by the type of ownership (i.e., for-profit or not-for-profit) and by demographic characteristics. We repeated the same analysis for the top five surgical procedures requiring an incision for both hospital-based and free-standing ASCs.

Emergency Departments. When physicians are unavailable or the complication originating in the ASC is severe, patients will often seek care from an Emergency Department (ED). For these analyses, we used the State Emergency Department Databases (SEDD) from AHRQ's HCUP family of databases for California, Maryland, and New Jersey. Specifically, we used the years 2006-2008 to be consistent with the period of time used for the SASD analyses. We identified all patients with infectious complications having their first encounter (without a diagnosis of infection) in the SASD and a successive encounter (with a diagnosis of infection) in the SEDD. Unfortunately, we identified very few surgical complications presenting to the EDs from ASCs, highlighting another limitation in identifying complications originating in the ambulatory surgery environment.

Hospitals. Hospitals provide a useful venue to further understand the scope and magnitude of the problem for two reasons. First, patients experiencing complications from an ASC may receive followup care for that complication in the hospital. Second, hospitals perform a large number of surgical procedures themselves and can, therefore, help inform the ambulatory environment, because greater than 80 percent of medical complications occur in the hospital setting. In addition, current administrative datasets provide extensive information on hospital encounters, patient characteristics, organizational structures, and resource utilization associated with each diagnosis.

For this study, we analyzed the Nationwide Inpatient Sample (NIS) databases for the States of California, Maryland, and New Jersey for the same period (2006-2008) to capture information about infection rates related to specific procedures. These infection rates were found to be below the infection rates reported in the literature for similar procedures, a result possibly due to underreporting and/or missing data points in the databases.

The study participants included all discharges in the SASD and NIS datasets. The proportion of discharges with a surgical complication with an ICD-9 (International Statistical Classification of Diseases and Related Health Problems) code was determined in each dataset. Next, specific patient and organizational characteristics were examined for their association with surgical complications. These characteristics broadly fall into the following categories:

  • Patient sociodemographic characteristics (age, gender, race, payer).
  • Health status and comorbidity (APR-DRG-defined severity and specific co-morbidities suggestive of chronic conditions).
  • Utilization (admission type, length of stay, and hospital characteristics—such as hospital bed size,** teaching status, urbanicity, ownership, and location).

The results of the data analyses for the California 2008 SASD are included in Appendix B and are being incorporated into the risk models, as appropriate.*** Of particular importance, additional targeted analyses were performed to assist with specific probability estimates to inform the risk models, as necessary, to ensure the comprehensiveness of the model. For example, if Hispanic patients were identified as a particular risk group from the literature, the proportion of Hispanic patients with an SSI would be analyzed and compared to non-Hispanic white and non-Hispanic black patients, to permit a more focused analysis of the risk as it relates to ethnicity and race.

Site Visits

The third source of process factor information comprised site visits to actual ASCs. AIR conducted site visits to four local centers to explore patient throughput variables of interest and to determine boundaries of the risk modeling exercise. Each site visit represented a different context in which SSIs can occur within the ambulatory surgery setting: an academic hospital-associated ASC; a community hospital-associated ASC; a free-standing, for-profit ASC; and a free-standing, hospital-associated pediatric ASC. Site visits were conducted between January and March 2011. Any differences in the context of care identified during the site visits helped inform the probabilistic risk assessment models about the risks for acquiring an SSI in these settings and created the opportunity for cross-fertilization and learning when these settings were compared. For example, a pediatric-specific ASC may have procedures to reduce risk in place for children undergoing surgery that can then be incorporated in ASCs that care for both children and adults.

To conduct the site visits, AIR prepared a semistructured interview protocol to use at each site, targeting questions based on the ASC staff member's roles and responsibilities. AIR submitted the protocol and methodology for review by AIR's Internal Review Board (IRB) and received approval on December 7, 2010. The protocol is presented in Appendix C.

For each site visit, the AIR team identified a key contact who would coordinate the visit for each organization. Site visits included three major activities that served as the basis for the sociotechnical element of the ST-PRA models:

  • A review and comparison of policies and procedures related to the occurrence and prevention of SSIs, including policies governing patient care, room cleaning, disinfection, and equipment disinfection and sterilization. This review also encompassed policies for procedures prior to surgery, during surgery, and after surgery.
  • Informal exploratory interviews with a selection of six staff, on average, from each participating ASC to learn about infection prevention policies and procedures in place.
  • Comparison of the process flow across sites, noting differences in policies and procedures, facility characteristics, and other relevant issues, as necessary.

Exhibit 4 identifies the different types of staff interviewed at each ASC. Specific names have not been included to ensure interviewee confidentiality, as established in the informed consent agreements. Please note that many site visit participants served in more than one role at the ASC.

Once the site visits had been completed, we created a series of tables for each surgical phase (i.e., preoperative, operative, and postoperative phases), to enable the comparison of policies, procedures, and practices across the four sites. In sum, interviewees discussed the steps involved from the preoperative call through patient discharge. Some of the similarities and differences found across the sites involved differences in the capacity and layout of the facility. For example, some ASCs are equipped with isolation bays reserved for patients with methicillin-resistant Staphylococcus aureus (MRSA) infections, whereas others are not. Although policies and procedures for handwashing were variable across the ASCs, the sites were consistent in requiring at least 15 seconds of handwashing between patients. The operative phase demonstrated the most consistency across study sites with respect to the surgical scrub, patient draping, and room preparation. The mapping of the typical patient's flow through the ASC, as well as facility structural factors, served as the foundation for building the process component of the fault tree model. The site visit comparison tables may be found in Appendix D.

Technical Expert Panel (TEP)

In addition to the other three data sources, AIR convened a panel of technical experts to guide the ST-PRA modeling. Members of the TEP were identified and selected to represent an array of expertise to ensure comprehensive coverage of the relevant issues. During the course of the project, AIR, in collaboration with AHRQ, added two additional experts to the panel, thereby increasing the comprehensive coverage of expertise represented on the TEP. Throughout the project, TEP input has helped shape the design of the fault tree model and the final intervention presented in this report. The members of the TEP, their specialties, and their affiliations are shown in Exhibit 5.

The initial TEP meeting was conducted on January 14, 2011. Participants at this meeting included the listed TEP members, except Drs. Martin and Song, who were subsequently added to the panel. In addition to the TEP members, Mr. David Marx from Outcome Engineering participated to provide ST-PRA modeling expertise. The purpose of this meeting was to orient TEP members to the project objectives and ST-PRA, gather feedback on the selection of one or more surgical procedures to serve as the focus for the ST-PRA, and gather feedback to be used for informing the initial fault tree development. Feedback from this meeting resulted in the specification of the study parameters referenced in the following section ("Resulting Decisions").

On June 6, 2011, the AIR team convened a second meeting by teleconference, using LiveMeeting. This meeting involved a subset of the original TEP (Dr. Ackerman and Ms. Crowley) and included two added TEP members, Drs. Song and Martin. The purpose of this meeting was to review the draft fault tree model and solicit feedback on areas for improvement. More details about this meeting are provided in the "Development of the Fault Tree Model" section of this report.

On November 2, 2011, the AIR team convened the third and final TEP meeting by teleconference using LiveMeeting. This meeting again involved a subset of the original TEP membership, including Drs. Aron, Coleman, Martin, Perz, Schwartz, Song, and Sorrentino, Ms. Crowley, and Mr. Wolf. The purpose of this meeting was to review the results of the ST-PRA modeling effort; to discuss the methods for identifying the basic events associated with the highest risk, as well as the unique combination of event sequences (cut sets) that lead to the occurrence of an SSI; and to obtain input on an intervention designed to reduce the likelihood of an SSI.

Resulting Decisions

Using the information gathered through the various sources (literature review, database analysis, site visits, and TEP input), the AIR team identified six important parameters to facilitate the development of the ST-PRA fault tree model. The parameters for the ST-PRA are as follows:

  • Examine only SSIs stemming from arthroscopy of the knee.
  • Examine only deep incision SSIs.
  • Limit the temporal period of interest from the preoperative call to 30 days after the procedure.
  • Focus specifically on procedures performed in free-standing and/or hospital-affiliated ASCs.
  • Develop a fault tree accounting for patients who present at risk in the ASC.
  • Develop a fault tree accounting for patients introduced to a microbe (or infection) in the ASC.

For more details on these parameters, please refer to Appendix E.

Return to Contents

 Development of the Fault Tree Model

A fault tree is a graphical depiction that conjoins risk estimates associated with a specific outcome of interest. For this study, the outcome of interest was an SSI occurring in outpatient arthroscopic surgery of the knee. The initial development of the fault tree used the four major inputs described above (i.e., literature review, database analysis, site visits, and technical expert input) to develop the risks associated with an SSI. Iteratively, the model was refined and revised to create a model that had face validity with technical experts who understand the procedure under study. In this section, we detail the steps taken in developing the fault tree model, as depicted in Exhibit 6.

Step 1. Identify All Factors Contributing to the Outcome of Interest

After determining the outcome of interest (i.e., in this case, the occurrence of an SSI, which is also the "top event" in the fault tree), the first step in constructing a fault tree involved identifying the risk factors (e.g., lack of communication between health care providers, patient does not comply with discharge instructions, failure to prepare skin appropriately prior to surgery, equipment failure) that are the most important contributors to this outcome. The end goal for this step was to identify a comprehensive list of variables (also known as "basic events") that contribute risk within the model and potentially lead to the outcome of interest. An initial list of basic events was created based on the major risk factors recognized in the extant literature as contributing to an SSI. This list was augmented by studying the process maps developed from the site visits and identifying probable points of failure in the processes (e.g., communication failure between health care professionals). Finally, based on discussions with TEP members in attendance at the June 6 meeting, some basic events were added to this list, and some basic events were removed because their contribution to the top event was considered negligible. When additional basic events were considered for inclusion in the fault tree, a targeted literature review using these basic events as key search terms was conducted to provide additional support for their inclusion.

Step 2. Identify the Dependencies and Interactions Among the Risk Points

Once the outcome of interest (i.e., top event) and all the basic events contributing to this top event were identified, the research team considered the ways the basic events were connected to the result in the top-level event. For many of the basic events, this process was straightforward. For example, it is clear that contamination contributes additional risk at the basic event level, which can lead to a higher frequency of SSIs. There are numerous ways to build contamination into different parts of the fault tree.

The approach we used to incorporate these risk points was derived by separating the basic events into components of the operative process: preoperative, operative, and postoperative. By creating a logic model and isolating basic-level events in each part of the operative process, we expected to improve the face validity and overall interpretability of the model by the clinicians participating on the TEP, as well as a broader audience, including hospital and ASC administrators, and other relevant stakeholders. It was also a useful method for incorporating the data gathered from the site visits, because contamination can occur with people, processes, or equipment, each of which can contribute independent risk for SSIs.

As a preliminary step, we established specific parameters to guide the development of the model framework and the relationships of the risk points. Parameters allow fault tree designers to home in specifically on a top-level event (i.e., SSI) and the characteristics of the setting where this event might take place. They also serve to guide subject matter experts about variables of interest and key top-level event characteristics relevant to fault tree design. For this project, we limited the procedure under study to arthroscopy of the knee, due to the high frequency of these procedures in the ASC environment. We limited the time frame under consideration from the preoperative call by the outpatient surgery center to 30 days postoperative, due to the higher rate of infection occurring within 30 days of surgery, using the definition of an SSI provided by the Centers for Disease Control and Prevention.6

Once the scope of the model was appropriately defined, the relationships (i.e., dependencies and interactions) among the multiple risk points were studied to understand how they collectively lead to an SSI. This is where clinical judgment, the results of the database analysis, site-visit process maps, and the input from the TEP were critical. Using these inputs enabled us to identify the multiple connections associated with the occurrence of SSIs. For example, a patient-level factor (e.g., diabetes) was identified from the literature, a staff-level factor (e.g., wearing artificial nails) was identified from a review of an ASC's policy, and an organizational-level factor (e.g., preoperative screening) was identified by the technical experts. Using multiple data sources was invaluable, because organizational-level factors and their connections to the top event may be specific to the different types of outpatient surgery centers. These connections were further enhanced by targeting additional literature searches on patient-level and staff-level factors, which have been the focus of many research studies.

 

The relationship between the basic events and the top event were established next. The fault tree uses "gates" to demonstrate the logic for joining all the basic events into an organized model that contributes to the outcome of interest. The two major types of gates are "AND" gates (i.e., the output event occurs if all input events connected to the AND gate occur) and "OR" gates (i.e., the output event occurs if at least one of the input events connected to the OR gate occurs). In combination, the basic events, modeled in the fault tree along with the AND gates and OR gates, produced a descriptive, hierarchical flow diagram of the process and the outcome under investigation. Exhibits 7 and 8 present examples of AND and OR gates, respectively.

Step 3. Validate the Fault Tree Model

Once an initial draft of a fault tree was developed, the model was reviewed by a subset of the TEP, as noted earlier, to obtain feedback on the connectivity and logic of the basic events with regard to the outcome (i.e., top event). The model was revised on the basis of this feedback. Leveraging additional contacts within the health care field, we addressed specific questions that needed further clarification through focused interviews with individuals knowledgeable about the outcome under investigation. Additional literature searches were conducted to address specific needs in the model related to risk points or probability estimates. The goal of this validation step was to confirm that the logical relationships built into the fault tree are representative of the real system and processes under study (e.g., the preoperative, operative, and postoperative processes for an arthroscopic knee surgery at the ASC). Appendix F (PDF) [ PDF file - 230.97 KB]  presents the final version of the fault tree.

Step 4. Identify the Likelihood of the Basic Events in the Fault Tree

The next step involved assigning probabilities (or likelihoods) to each basic event in the fault tree. To the extent that it was available, information from the peer-reviewed and grey literature was used to provide a starting point for estimating the probabilities of the basic events. When we found gaps in these estimates, we performed additional and more focused literature reviews or interviews with knowledgeable individuals to derive the estimates for these probabilities. It is important to note that when we needed to rely on technical experts' estimates, we targeted these relationships in the subsequent sensitivity-testing component of this study. Appendix G presents the references for the probabilities used in this step.

Once the probabilities were assigned for the basic events, the fault tree was modeled using Relex, a software package that calculates the remaining probability estimates for all intermediate and top-level events in the fault tree using the logical relationships (e.g., AND gates, OR gates) previously specified. For the AND gates, the probabilities of its input events are multiplied together; for the OR gates, the probabilities of its input events are added together, with the overlap subtracted to prevent double counting of the gate if both failures occur simultaneously. This process leads to a probability of occurrence of the top-level event, along with the major risk points in the process (also known as cut sets) that were developed as the next step of this project.

Step 5. Conduct Sensitivity Analyses of the Fault Tree Model

Because some probabilities included in the fault tree model were based on imprecise information from the available databases, highly variable risk estimates in the literature, or estimates from technical experts, we conducted a series of sensitivity analyses to improve the reliability of the modeling exercise. The sensitivity analysis can be considered a series of grounded "what if" tests to study the robustness of the ST-PRA model. These analyses began with an examination of the base case, which corresponds to the current fault tree model detailed in the interim report, and then varying the basic event probabilities across a range of values to determine whether the combinations of the major events cause a change in the risk of an SSI. These analyses involved identifying the minimal cut sets, defined in the next section, for the base case and for each variation of the base case (obtained by modifying the probabilities) to study the robustness of the fault tree model. This process allowed the team to identify an intervention that would have the greatest likelihood of mitigating the risk of SSIs, a major goal of this project.

Minimal cut sets. Cut sets are a unique combination of events leading to the occurrence of a top level event (an SSI). A cut set is considered a minimal cut set if, when any basic event is removed from the set, the remaining events are collectively no longer a cut set. A minimal cut set is defined as a critical path through multiple failure points.7 By identifying the different cut sets associated with an event, the model can be reconsidered after removing specific failure points or system components as a result of implementing an intervention or series of interventions designed to reduce the rate of occurrence of the top event. The minimal cut sets are identified through the software, using the underlying logic as depicted in the AND/OR gates. The software then combines basic level event probabilities to identify the paths, based on the conditional probabilities of event combinations. The minimal cut sets with the highest risk for the top level event are then listed in descending order of priority.

Sensitivity analysis. In the sensitivity analyses, we focused on the events for which the literature reported large variations in the probabilities, and varied these probabilities in the base case within the ranges suggested in the literature. When a probability estimate was not available from the literature, an anchor estimate was obtained from technical experts in the field. For example, when questions arose about the likelihood of a failure in the process with relevance to pediatric patients, Drs. Song and Coleman, both members of the TEP, were asked to estimate the probability of this risk occurring, based on their professional experience. This estimate was then considered the anchor estimate for the sensitivity analyses, which examined the range of intervals from 25-75 percent around the provided probability estimate.

For example, handwashing is a common approach for helping to prevent the spread of bacteria and would thereby be expected to have a positive impact on preventing the occurrence of SSIs. The literature indicates that non-OR staff compliance rates for handwashing range between 40 and 90 percent. The OR staff compliance rates for handwashing are reported to be consistently higher and with a much lower variation, around 75-90 percent. In the sensitivity analysis, we varied the conditional probability for non-OR handwashing compliance across the range of 40-90 percent to better understand the impact that handwashing may have at mitigating the occurrence of the SSI. The sensitivity analyses ensured that the model was appropriate even if the probabilities of basic level events constituting the model were grossly inaccurate at the beginning of the modeling exercise. If the same contributors are identified after the sensitivity analyses, the model's integrity can be ensured. If this does not happen, further data would need to be collected (e.g., through additional interviews with the health care providers) to increase the reliability of the probability estimates.

 We ran the fault tree model for each variation of the base case and determined the corresponding SSI rate and the top five minimal cut sets to understand how and if they had changed over the base case. Exhibit 9 presents the top five minimal cut sets for the base case with their contribution to an SSI (labeled Contributed Probability). Exhibit 10 displays the probability variations considered in the sensitivity analyses.

Sensitivity analyses were run across variations of the base case, each corresponding to a change in probability of an event, as defined in Exhibit 10. Each of the resulting minimal cut sets included the following common events:

  • Event 173 Staff fails to provide patient with instructions for weight reduction.
  • Event 660 Patient fails to notice infection during home care.
  • Event 642 Staff fail to protect patient effectively.
  • Event 543 SSI risk for obese patient, weight not reduced and nutrition not improved.
  • Event 450 Obese, but not diabetic, patient (30 ≤BMI <40).
  • Event 182 Fail to administer indicated antibiotics.

The events overlapping these minimal cut sets provide the biggest opportunity for preventing risk. As a result, these five events became the focus of our next step, designing a risk-informed intervention, which is presented in the following chapter.


* Note that the New Jersey SASD did not contain data on free-standing ASCs, whereas both the Maryland and California SASDs did. However, the California SASD included more variables that were of interest to this study and also included many more records than the Maryland database.
** That is, the number of beds in use or available for use.
*** The results for the California 2006 and 2007 SASD are similar to those of the 2008 database and have not been included in Appendix B.

Page last reviewed April 2013
Internet Citation: Chapter 2. ST-PRA Development: Proactive Risk Assessment of Surgical Site Infection in Ambulatory Surgery Centers. April 2013. Agency for Healthcare Research and Quality, Rockville, MD. http://www.ahrq.gov/research/findings/final-reports/stpra/stpra2.html