3 Methods

3.1 Reporting Requirements

The 2007 amendment to the MCARE Act (Act 52) requires all hospitals to report all HAIs into CDC’s National Healthcare Safety Network41. This report contains data from inpatient hospitals and only includes patients that had overnight hospital stays. Facilities are required to follow the NHSN Patient Safety Protocol to identify, classify, and report infections and patients at risk. Act 52 also requires facilities to have an HAI surveillance plan to monitor HAIs. Case finding and case classification are the responsibility of the facility.

The Department released guidance for reporting of C. difficile and MRSA infections that permits certain types of hospitals to report these infections in one of two ways. Acute care hospitals, long-term acute care hospitals and inpatient rehabilitation facilities are required to report positive laboratory results for C. difficile and MRSA found in the blood. These are called LabID events. Psychiatric, children’s hospitals, and critical access hospitals are not required to report LabID events; therefore, data from these hospital types are not included in this report. These facilities have an option to report either LabID events or MRSA bloodstream infections and CDIs that meet a surveillance infection definition. These infection surveillance definitions require both clinical symptoms and a positive laboratory test, whereas LabID events only require a positive laboratory test.

The descriptions below summarize procedures and case definitions that are extensively detailed in the 2018 NHSN Patient Safety Protocol, which is the authoritative source for the classification of HAIs included in this report.41

3.2 Benchmarked Infections

This report includes device-associated, procedure-associated HAIs and LabID events. The CDC calculates SIRs based on the HAI and event data. HAIs included in this report are CLABSIs, CAUTIs, seven procedure-related SSIs (hip prothesis and knee prothesis, abdominal hysterectomies, colon surgeries, cardiac surgeries, and cardiac bypass surgeries with one or two incisions), MRSA bloodstream LabID events, and CDI LabID events.

3.3 Reporting Period

This report includes infections that occurred in 2018 and 2017. For hospitals with less than one predicted infection during 2018, two years of data are combined to form statistical metrics. For hospitals with one or more predicted HAIs during 2018, only 2018 data are presented. The report also includes SSIs resulting from procedures performed in late 2018 that had an infection diagnosed in 201941 per NHSN timing criteria.

The 2018 NHSN analysis datasets were downloaded on June 11, 2019. The 2017 CLABSI, CAUTI and SSI data were locked down on November 15, 2018.

3.4 Included Facilities

This report includes data from acute care, women’s, children’s, long-term acute care, rehabilitation, psychiatric, and critical access hospitals. Free-standing rehabilitation hospitals and inpatient rehabilitation units in hospitals are classified as inpatient rehabilitation facilities. Facility data was excluded from this report if the facility was not open for all 12 months of 2018.

3.5 Data Validation

The NHSN online data collection tool has several automated data validation algorithms. When inconsistent or out-of-range data are entered, an alert is displayed with a message prompting the user to correct the data value. In addition to the built-in NHSN tool, the Department analyzes facility data quarterly to identify missing, duplicative, or unusual data. The Department provides each facility with a facility-specific list of data elements to correct. Facilities have 30 days from the end of the analysis month to correct their data.

3.6 CLABSIs

The NHSN Patient Safety Component Manual41 defines a CLABSI as a lab confirmed infection of the blood in a patient who had had a central line for three or more consecutive days. The catheter must have been in place on the day that the infection was identified or the day before. Inpatients that receive dialysis through a central line are at risk for a CLABSI. A blood stream infection in a patient with a central line that is secondary to a recently documented infection at another body site, with the same organism, is not classified as a CLABSI. The number of CLABSIs are aggregated monthly by hospital location (unit) and entered into NHSN each month. Patient information is not reported.

Hospitals are not required to follow-up with discharged patients to conduct infection surveillance. However, infections that are diagnosed after discharge are counted if they are reported to the hospital and meet the NHSN criteria.

3.7 CAUTIs

The NHSN Patient Safety Component Manual41 defines a CAUTI as lab confirmed infection in the urine from a patient with at least one symptom who had had a urinary catheter for three or more consecutive days. The catheter must have been in place on the day that the infection was identified or the day before. An infection is defined as bacterial growth of two or fewer species at least one of which is heavy growth (i.e., more than 100,000 colony forming unit per milliliter urine). Repeat infections are infections at the same site that return within 14 days of the initial date of onset; these are not counted again. The number of CAUTIs are aggregated monthly by hospital location (unit) and entered into NHSN each month. Patient information is not reported. For more information refer to these resources Frequently Asked Questions about Catheter-associated Urinary Tract Infections.

Hospitals are not required to follow-up with discharged patients to conduct infection surveillance. However, infections that get diagnosed after discharge are counted if they are reported to the hospital and meet the NHSN criteria.

3.8 Counting Device-Days, Patient-Days, and Admissions

Because CAUTIs and CLABSIs only occur among people who had an indwelling catheter in place, it is important to track and report the number of patients that have an indwelling catheter each day. These patients are at increased risk of developing an HAI. Hospitals with many device-days (see below for definition) are predicted to have more infections than hospitals with fewer numbers of device days.

The number of persons-at-risk is measured as urinary catheter days or central line days (device days). They are defined as follows:

  • Urinary catheter days: the total number of hospitalized patients with a urinary catheter in place per day over the month (total number of patients with a urinary catheter in place multiplied by the number of days a catheter was used for each patient).

  • Central line days: the total number of hospitalized patients with a central line in place per day over the month (total number of patients with at least one central line in place multiplied by the number of days a central line was in place for each patient).

The process of counting the total number of at-risk patients can be performed in one of several ways. Hospitals can perform a daily census by hand, or a once a week manual count that is extrapolated to represent the month, or through use of a counting algorithm in the hospital’s electronic medical record. Counting is performed separately for each hospital location (e.g., oncology ICU, pediatric unit, burn unit) and entered each month into NHSN.

The number of newly hospitalized patients is counted every day or week using similar processes as those described for counting device days. This is performed separately for each hospital location (e.g., oncology ICU, pediatric unit, burn unit). The counts are aggregated by location and month and entered into NHSN each month.

3.9 Surgical Site Infections

3.9.1 Counting Surgeries

Each surgical procedure that occurred in calendar year 2018 and met the reporting criteria described in Chapter nine of the Patient Safety Component Manual41 was eligible for inclusion in this report. Only surgeries for which the patient was in the same hospital overnight are included in this report. As per NHSN guidelines, infections with well-known community-associated organisms and/or organisms associated with latent infections were not classified as SSIs41. This report includes SSI data from the below seven most common surgeries:

  • Cardiac (CARD);
  • Cardiac bypass graft surgery with one incision (CBG with one incision);
  • Cardiac bypass graft surgery with two incisions (CBG with two incisions);
  • Knee prosthesis (KPRO);
  • Hip prosthesis (HPRO);
  • Abdominal hysterectomy (HYST); and
  • Colon surgery (COLO).

Since diabetes is an important risk factor for infection and should be accounted for in statistical models, surgical patients in whom the diagnosis of diabetes was undetermined were excluded. Similarly, surgical patients who received insulin for perioperative control of hyperglycemia but had no diagnosis of diabetes were also excluded.

3.9.2 Counting SSIs

The NHSN Patient Safety Component Manual requires hospitals to use active, patient-based, prospective surveillance to identify superficial incisional, deep incisional, and organ space SSIs41. Post-discharge and before-discharge surveillance methods might include:

  1. Direct examination of patients’ wounds in the hospital or during follow-up visits to either surgery clinics or physicians’ offices;
  2. Review of medical records or surgery clinic patient records;
  3. Surgeon surveys by mail or telephone; and
  4. Patient surveys by mail or telephone.

An SSI is classified as being either superficial incisional, deep incisional, or organ space. Superficial and deep incisional infections are further classified into those identified at the primary or secondary incision site among surgeries that had two or more incisions.

A superficial incisional infection involves only the skin or subcutaneous tissue, whereas a deep incisional infection involves deep soft tissues of the incision41. Superficial incisional infections must have one of the following:

  1. Purulent drainage;
  2. Identification of organism from wound;
  3. Reopening of the incision by the provider, without obtaining a culture, and clinical symptoms consistent with infection; or
  4. Diagnosis of a superficial incisional SSI by the surgeon or other designee.

Deep incisional SSIs involve deep soft tissues of the incision (e.g., fascial and muscle layers), and the patient must have at least one of the following:

  1. Purulent drainage from the deep incision;
  2. A deep incision that spontaneously dehisces, or is deliberately opened or aspirated by a surgeon or other designee and organism identified by a culture or non-culture-based test; if culture or non-culture testing is not performed, then the patient must have consistent symptoms of infection (i.e., objective fever or localized pain or tenderness); or
  3. An abscess involving the deep incision that is detected on gross anatomical, histopathologic exam or imaging test.

An organ space infection occurs when tissue deeper than fascial/muscle layers is infected. NHSN recognized 14 different types of organ space infections. Organ space infections that may arise from any of the three cardiac surgeries include osteomyelitis, myocarditis or pericarditis, endocarditis, intraabdominal, lung, mediastinitis, and arterial or venous infection. Intraabdominal, deep pelvic tissue infection or other infection of the male or female reproductive tract, or urinary system infection may result from colon surgery. Hip or knee replacement surgery may result in osteomyelitis or periprosthetic joint organ space infections. Intraabdominal, or deep pelvic tissue infection or other infection of the female reproductive tract or vaginal cuff infection could result from an abdominal hysterectomy.

Irrespective of the type of surgical procedure, superficial SSIs must occur within 30 days of the procedure. The surveillance period for deep incisional and organ space SSIs varies by procedure type. Patients who had a COLO or HYST procedure were monitored for 30 days, whereas others were monitored for 90 days.

There are two different methods to examine SSIs. In one method, SSIs that affect any part of the superficial, deep, and organ space are counted. In the other method, only infections located in deep tissues or the organ space are counted. These infections are more serious and easier to identify. To address these different methods, NHSN calculates two SIRs, one for each SSI classification method. The “all” method includes superficial, deep incisional, and organ space infections, whereas the “complex” method includes only deep tissue and organ space infections.

3.10 Laboratory Identified (LabID) Events

It is less labor intensive to use positive lab tests alone to identify and report infections compared to methods which require both clinical criteria and lab tests. It is for this reason that Clostridioides difficile infections (CDI) in the stool and methicillin-resistant Staphylococcus aureus (MRSA) from the blood are reported as LabID events. This simple case definition ensures that the reported data are standardized and consistent between facilities. LabID events that are from clinical, not surveillance, specimens are required to be reported. Unlike the four other HAIs, both inpatient and outpatient results (i.e., from the emergency department and 24-hour observation locations) are reported. These data can be used to estimate the infection burden in the inpatient setting, the outpatient setting, and the community. LabID events are used by CMS for quality reporting programs.

3.10.1 Patient Days, Admissions, and Encounters

Facilities report a monthly count of patient days (like device days) and admissions for all inpatient locations separately regardless of admission and/or billing status. Facilities exclude patients in inpatient rehabilitation facility (IRF) locations and inpatient psychiatric facility (IPF) locations with a unique CMS Certification Number (CCN) from monthly facility counts. Admissions and discharges from hospital locations to either IRF/IPF and vice versa are considered “transfers.”

Separate counts of patient visits to the emergency department and 24-hour observation unit are also reported monthly. These visits are called “encounters.”

3.10.2 Classification of Positive Lab Events

NHSN applies an automated process to the LabID event data to classify the events as either healthcare-facility onset or community-onset. Healthcare-facility onset cases occur on or after the fourth day following the admission date (where the date of admission is the first day). Community-onset cases occur among patients admitted to the hospital with a positive test on days one, two, or three and among all patients with a positive test who visit the 24-hour observation unit or emergency department affiliated with the hospital. An event is classified as an incident if the patient never had a prior positive C. difficile test from the same hospital or if the previous positive test was 57 or more days prior to the second test date. Only incident, healthcare-facility onset LabID events are reported in this report. Community-onset events and respective denominator data are used for risk adjustment in the statistical models, discussed later.

3.10.3 MRSA LabID events

Healthcare-facility onset MRSA bloodstream infections are defined as a positive test for MRSA from blood that meet the above criteria. There are two instances in which a second positive MRSA bloodstream test collected after a hospitalization is counted as a second healthcare-facility onset infection from the same patient: If the specimen is collected after the patient moved to a new hospital location or if the specimen is collected 14 or more days after the first collection date. If a hospitalized patient has a second positive test after being discharged, this is classified as community-onset event.

3.10.4 CDI LabID events

Tracking and reporting C. difficile infections are difficult for three main reasons. First, C. difficile can live in the intestinal flora of people and not produce the toxins that cause clinical symptoms. These individuals are considered “colonized” and not infected with C. difficile. Second, CDI is difficult to treat and often reoccurs. This makes classifying recurrent and new onset cases difficult. Third, diarrhea in hospitalized patients is common, and C. difficile is the cause in less than 30% of cases. In this report, only incident, or first time, healthcare facility onset events that meet the above criteria are counted as a CDI LabID event.28.

To ensure that positive test results reflect true infections, hospitals are advised to test only appropriate stool samples for CDI. The samples should conform to the shape of the container and have a fluid-like consistency. The patient should have three or more diarrhea episodes within a 24-hour period that meet this criterion.

Because healthy babies can carry C. difficile in their intestinal flora, results from neonatal ICUs, special care nurseries, and labor, delivery, recovery, and postpartum units, and well-baby nurseries are excluded from the monthly admission and patient day monthly counts. Specimens from outpatient well-baby clinics are excluded from the monthly encounter counts.

Patients seen in hospital-affiliated outpatient locations (not just the emergency department and 24-hour observation area where care is provided post discharge or prior to admission) are counted as encounters and eligible to be reported. These cases are classified as community-onset cases.

Microbiology labs can choose between four different categories of CDI tests, each of which has different sensitivities (ability to identify a true positive) and specificities (ability to identify a true negative). Two tests detect the presence of C. difficile toxin. The cell cytotoxicity assay (ToxiCulture) is positive when stool added to a monolayer of an appropriate cell line produces a cytotoxic effect. The second test, in which both stool and neutralizing antitoxin are added, does not produce a cytotoxic effect. The enzyme immunoassay (EIA) test uses monoclonal antibodies to detect C. difficile toxin A and polyclonal antibodies to detect C. difficile toxin B. The EIA method is used to detect the glutamate dehydrogenase (GDH) antigen, which is specific to C. difficile (but not specific to a C. difficile toxin). The nucleic acid amplification test (NAAT) detects genes that make the C. difficile toxin. However, NAAT cannot determine whether the toxin genes are turned on or off (to produce toxin).

Experts have not reached consensus on an optimal lab method to detect an active CDI case. There is a growing trend to use two laboratory tests, one to detect a toxin and the other to identify the organism. When this two-step process is used, the finding of the second or last test result shall determine whether the specimen is positive or not. Most labs only use the results of a single test to determine whether the specimen is positive. Each facility reports the type of CDI lab test method most used during each quarter.

Duplicate positives CDI tests are not reported and are defined in the 2018 Patient Safety Component41 as any C. difficile positive laboratory result from the same patient and hospital location/unit, following a previous C. difficile positive laboratory result within the past 14 days (even across calendar months and readmissions to the same facility). There should be 14 days with no C. difficile positive laboratory result for the patient and location/unit before an additional C. difficile LabID event is entered into NHSN for the patient and location. The date of specimen collection is considered day one.

3.11 Data Analysis

Statistical metrics were calculated in NHSN to evaluate hospital and state performance. These metrics compare Pennsylvanian HAI data to 2015 data from similar hospitals from the rest of the country. This reference group is referred to as the 2015 baseline data. This method of comparison and calculation enables the metrics to be compared from year-to-year.

3.11.1 Standardized Measures of Performance

NHSN calculates two metrics that are reported in this report: the standardized infection ratio (SIR) and the standardized utilization ratio (SUR). Both metrics quantify the degree to which the actual number of HAIs or device days differ from what occurred in the 2015 baseline data.

The SIR measures HAIs, whereas the SUR measures either central line days or urinary catheter days. The statistical methods used to calculate both the SIR and SUR are virtually identical and the process is described below. The word “SIR” is used in the below description for the sake of consistency, although, the same basic method applies to creation and interpretation of the SUR. The differences between SIR and SUR will be discussed when applicable.

3.11.2 Interpreting SIRs and SURs

In this data analysis, the SIR is the ratio of the actual (observed) number of HAIs reported compared to the number that would be predicted based on the 2015 national data, adjusting for risk factors that have been found to be significantly associated with the infection incidence. The SIR is calculated by dividing the number of observed infections by the number of predicted infections.

The statistical model adjusts for risk factors that vary among hospitals that may underlie differences in the number of reported infections. When these types of statistical models are used, the data are considered “adjusted” or “standardized.” The predicted number of infections is derived from the statistical model and is interpreted as the number of infections a nearly identical hospital would have during 2015. When comparing two SIRs, those factors are unlikely to explain why hospitals with lower SIRs are doing better than those with higher SIRs. The differences are due to other causes, such as adherence to infection prevention and control practices or patient characteristics.

The SIR and SUR are useful metrics to compare one facility to similar facilities in the rest of the country. Below are rules to interpret SIRs by themselves:

  • If the SIR > 1.0, then more HAIs were observed than predicted, based on the 2015 national aggregate data.
  • If the SIR= 1.0, then the same number of HAIs were observed as predicted, based on the 2015 national aggregate data.
  • If the SIR < 1.0, then fewer HAIs were observed than predicted, based on the 2015 national aggregate data.

SIRs do not estimate the chance, or risk, of getting an HAI. It only compares the actual number of infections compared to the number predicted by the 2015 baseline data. A few problems arise when using the SIR as an indicator of HAI risk. The biggest problem is that hospitals with few patients and/or beds are not predicted to have 1.0 HAI. CDC and statisticians agree that SIRs among facilities with less than 1.0 predicted infection are not an accurate estimate of the infection risk. Calculation of a SIR in these hospitals could result in a very large SIR value if the facility reported even one infection, and this may be misleading. In these cases, CDC recommends that a SIR be calculated over a two-year time span. This optimizes the estimation of the SIR because the number of observed and predicted infections will increase. After combining 2017 and 2018 surveillance data, some hospitals still did not have the requisite predicted infections of 1.0 or more. For these hospitals, the SIR was not calculated because the predicted number of infections was less than 1.0. This problem does not occur for standardized utilization ratios (SURs) because the predicted number of patients with a catheter was never less than 1.0.

Another problem is that low values of predicted infections lead to SIRs that are unstable. To further evaluate the SIRs value, it’s important to consider if it is statistically meaningful. For a small hospital that performs few operative procedures, the level of confidence in the stability of the SIR is lower than for a hospital with many procedures. This is reflected in a calculation known as the confidence interval (CI), for which a lower value and an upper value are calculated and displayed. The true SIR for a hospital is found somewhere between the lower value and the upper value with 95% certainty. In general, the smaller the facility, the wider the confidence interval, meaning that confidence in the calculated measure is low. Larger facilities tend to have narrower confidence intervals, meaning the true value is likely to be in a tighter range of values.

To fully interpret SIRs, it’s best to examine not only how large or small the SIR value is when compared to 1.0, but also examine the 95% confidence interval. If a hospital has a 95% confidence interval that does not include 1.0, it is considered “statistically significant.” The combination of both approaches is considered the best way to evaluate a single SIR. Below are rules for interpreting the 95% confidence interval.

95% Confidence Interval:

  • The 95% confidence interval is a range of values in which a high degree of confidence exists that the true SIR (or SUR) lies within that range.
  • If the confidence interval does not include 1.0, then the SIR is significantly different than 1.0 (i.e., the number of observed infections is significantly different than the number predicted).
    • Example: 95% confidence interval = (0.85, 0.92)
  • If the confidence interval includes the value of 1.0, then the SIR is not significantly different than 1.0 (i.e., the number of observed infections is not significantly different than the number predicted).
    • Example: 95% confidence interval = (0.85, 1.24)
  • If the SIR is 0.000 (i.e., the infection count is 0 and the number of predicted infections is >= 1.0), the lower bound of the 95% confidence interval will not be calculated.
    • Example: 95% confidence interval = (0, 1.49)

If a hospital reported 10 CAUTIs during 2018 and only five CAUTIs were predicted in that hospital, the SIR would be 10 divided by five or 2.00. If another hospital reported five CAUTIs and 10 were predicted, the SIR would be five divided by 10 or 0.50 and the 95% confidence interval is calculated as 0.183 - 1.108. This means that, although the hospital appears to have half of the predicted infections, it’s 95% likely that this has just as many CAUTIs as a similar hospital. Because the value of 1.0 is within the 95% confidence interval, the SIR is not statistically significant at the 0.05 level.

3.11.3 CLABSI Statistical Models

Negative binomial regression was used to adjust for risk factors and estimate the number of predicted CLABSIs42. The strongest risk factor associated with CLABSIs is type of hospital. Therefore, separate models were created for each type of facility: acute care hospitals (ACH), critical access hospitals (CAH), long-term acute care (LTAC) facilities, neonatal intensive care units (NICUs), and inpatient rehabilitation facilities (IRF). A separate model for children’s, women’s, and psychiatric hospitals was not necessary, and those hospitals were included in the ACH model. A separate model was also used among patients in the neonatal intensive care unit (NICU). Twenty-two hospitals reported data from this unit location.

SIRs should only be compared over time within the same hospital or reporting group. Although not ideal, the observed and predicted CLABSIs from NICUs were added to those from the ACH model to create a single SIR or SUR for hospitals with NICUs.

The ACH CLABSI model adjusted for number of device days, number of beds, affiliation with a medical school, and type of unit. Risk factors incorporated in other hospital models for other hospital types included patient population characteristics such as proportion of patients admitted with a stroke. Please refer to the SIR Guide Supplement for a list of all models and risk factors included in each model42.

3.11.4 CAUTI Statistical Models

Negative binomial regression was also used to adjust for risk factors and estimate the number of predicted CAUTIs42. Similar to CLABSI models, separate models were used for each type of facility: ACH, CAH, LTAC facility, and IRF. Children’s, women’s, and psychiatric hospitals were included in the ACH model. A separate NICU model was not necessary.

SIRs should only be compared over time within the same hospital or reporting group.

The ACH CAUTI model adjusted for number of device days, number of beds, affiliation with a medical school, and type of unit. Risk factors incorporated in other hospital type models included patient population characteristics such as average length of stay. Please refer to the SIR Guide Supplement for a list of all models and risk factors included in each model42.

3.11.5 SSI Statistical Models

Logistic regression models were used to estimate predicted SSIs42. Models were built for each SSI separately. The SSI SIR was only calculated for ACHs and CAHs. Because age group was a risk factor for SSIs, each surgical procedure type had a separate model for adult and pediatric patients.

Only risk factors that were significantly related to SSIs were included in each of the 14 models, two models for each of the seven surgical procedures. Generally, the statistical models adjusted for patient-specific characteristics. Some models also adjusted for medical school affiliation and hospital size. The observed and predicted SSIs from adult and pediatric patients were summed at the hospital level to calculate the single facility SIR.

The statistical process used to calculate the predicted number of infections required the presence of all risk factor data. Procedures without risk factor data were not included in the analysis. Procedures with risk factor data that were substantially more or less than the average (and likely caused by data entry problems) were also excluded from the analyses. Very few Pennsylvania procedures were excluded for these reasons due to the validation process described previously.

As mentioned in the section entitled “Counting SSIs”, NHSN calculates SIRs for two different definitions of SSIs. In the “all” model, superficial, deep tissue, and organ space infections are counted. In the “complex” model, only deep tissue and organ space infections are counted as SSIs. CDC published national and state-specific SSI SIRs using the complex model in their 2018 HAI Progress report37. To facilitate comparison with these data, the Results section of this report defines SSIs using the same complex model. Please refer to the SIR Guide Supplement for a list of all models and risk factors included in each model42.

Journal articles that report SSIs to assess the magnitude of the SSI problem count infections from both superficial, deep tissue, and organ space sites.19 ,18 To evaluate these SSIs among Pennsylvania hospitals, Appendix A contains SIRs from the “all” model. The reader is encouraged to carefully interpret SSI SIRs from both the Results section and Appendix A. SSI SIRs can only be compared with 2015 or later NHSN data that use the same adjustment model (“all” or “complex”).

3.11.6 MRSA Bloodstream Infection Statistical Models

A negative binomial regression model was used to adjust for risk factors and estimate the number of predicted MRSA blood LabID events42. Separate models were used for each type of facility: ACH, CAH, LTAC facilities, and IRF. Children’s, women’s, and psychiatric hospitals were included in the ACH model. It is not recommended to compare SIRs between hospitals. SIRs should only be compared over time within the same hospital or reporting group.

The ACH MRSA LabID model adjusted for patient days at risk, both inpatient and outpatient community-onset prevalence rate, average length of stay, number of ICU beds, and affiliation with a medical school. Information about unit was not collected, so it could not be included in the model. No adjustment was made for CAHs or IRFs. Events in LTAC facilities were adjusted for percent of admissions on a ventilator. Please refer to the SIR Guide Supplement for a list of all models and risk factors included in each model42.

3.11.7 CDI Statistical Models

A negative binomial regression model was used to adjust for risk factors and estimate the number of predicted CDI LabID events42. Separate models were used for each type of facility: ACH, CAH, LTAC facilities, and IRFs. Children’s, women’s, and psychiatric hospitals were included in the ACH model. It is not recommended to compare SIRs between hospitals. SIRs should only be compared over time within the same hospital or reporting group.

The regression models were applied to quarterly data to determine the number of predicted cases. The quarterly data were added together to compute the annual SIR presented in this report.

Some CDI statistical models adjusted for type of CDI lab test by grouping the different one-step and two-step test methods into three categories. Those that used EIA for toxin, or GDH and EIA for toxin (two-step) or NATT and EIA (two-step) were in one group. Those that used EIA for toxin or GDH antigen and EIA for toxin (two-step), or NAAT and EIA (two-step) were in the second group. The remaining test types were placed in the third group42.

The ACH CDI model adjusted for the following factors:

  • patient days at risk;
  • inpatient community-onset CDI prevalence rate;
  • number of ICU beds;
  • type of CDI test;
  • number of licensed beds in the facility;
  • whether the hospital reported data from the 24-hour observation unit and emergency department (or not);
  • whether the hospital is a teaching hospital (or not);
  • whether the hospital was a specialty, non-specialty or oncology hospital (or not); and
  • presence of an affiliation with a medical school (or not).

Information about unit was not collected, so it was not included in the model. The CAH model adjusted for inpatient community-onset CDI prevalence rate and patient days at risk. The LTAC facility model adjusted for inpatient community-onset CDI prevalence rate, type of CDI test, percent of patients on a ventilator, percent of rooms that are single occupancy and patient days at risk.

The IRF model adjusted for type of CDI test, whether the facility/unit had more than 23.9% of admissions with orthopedic conditions, whether the facility/unit had greater than 5.2% of admissions with spinal cord disfunction, whether the facility/unit had more than 23.8% of admissions with a stroke and patient days at risk. The model also accounted for whether the facility was in a hospital or free standing, and, if free standing, whether community-onset CDI events were reported (or not). The SIR Guide Supplement lists all models and more details about the risk adjustment for in each model42.

3.12 Limitations

Comparisons of facility level SIRs from this report and the Pennsylvania 2016 and 2017 HAI reports are encouraged. However, comparisons with Pennsylvanian HAI reports prior to 2016 are not recommended because the methods used to calculate the SIR used different comparison hospitals. Additionally, because the definitions of HAIs changed over time, it is not recommended to compare case counts between this report and those prior to 2016.

Comparisons of SIRs between hospitals is discouraged. The statistical adjustment could not account for differences between patients that are related to risk of getting an infection (such as diabetes status). Patient characteristics were not collected for CLABSI and CAUTIs. The models cannot account for differences between hospitals, such as patient characteristics of those admitted to the hospital. Remnants of patient or hospital characteristics, such as measures of comorbidity, physician experience, and patient age, may account for the differences in SIRs that are observed between two different hospitals.