Study Design

For this study, we recruited a prospective cohort of general medical and surgical patients at the University of Maryland Medical Center, an academic tertiary care hospital. The patients in this study were initially recruited between January 2010 and November 2010 for a prospective study of the relationship between contact precautions and depression, anxiety, and patient satisfaction.Reference Day, Perencevich and Harris ^{12 ,} Reference Mehrotra, Croft, Day, Perencevich, Pineles, Harris, Weingart and Morgan ¹³ Patients on contact precautions were individually matched with patients not on contact precautions by a minimum 3-day hospital length of stay and by admission service (ie, to limit bias related to patients on contact precautions having longer length of stay and more severe illness). Study participants on contact precautions remained on contact precautions for the duration of their hospital stay, and patients not on contact precautions also did not change contact precautions status during the study period. This study was approved by the Institutional Review Board of the University of Maryland, Baltimore.

Measurements and Outcomes

Demographic data such as age, gender, education, minority status (Hispanic and/or non-white race), and marital status were collected at study enrollment; Charlson comorbidity score and length of stay were obtained from the hospital central data repository. We defined adverse events as “unintended physical injury resulting from or contributed to by medical care that requires additional monitoring, treatment, or hospitalization or that results in death” as used by the Institute for Healthcare Improvement (IHI) Global Trigger Tool.Reference Griffin and Resar ² We did not consider events due to the patient’s underlying condition as adverse events. Adverse events were identified using the IHI Global Trigger Tool, which is a standardized method for completion of structured medical record reviews within approximately 20 minutes.Reference Griffin and Resar ² The trigger tool detects more adverse events than approaches such as the Agency for Healthcare Research and Quality (AHRQ) Patient Safety Indicators or hospital reporting systems.Reference Naessens, O’Byrne, Johnson, Vansuch, McGlone and Huddleston ^{14 ,} Reference Classen, Resar, Griffin and Federico ¹⁵ Three initial reviewers used the trigger tool to review the patient charts of all members of the cohort for adverse events. One initial reviewer (LDC) reviewed the records of all 296 cohort members, and 2 initial physician reviewers (ML and JL) each reviewed half (N=148) of the cohort. A final physician reviewer (DJM) adjudicated any disagreements on adverse event occurrence between initial reviewers, per the IHI Global Trigger Tool method.Reference Griffin and Resar ² All reviewers received training on using the IHI Global Trigger Tool and practiced with standardized sample charts, following the training approach used by the tool developers.Reference Harris, Pineles and Belton ¹⁶ Reviewers determined whether each adverse event initially began before, during, or after the admission of interest. Analysis was restricted to adverse events that began during the index admission only, as events that began prior to or after admission may not have been the result of the use of contact precautions.

Our primary outcomes were noninfectious and infectious adverse event rates considered separately, and the secondary study outcomes were preventable and severe noninfectious adverse events. Infectious adverse events were defined as infections that occurred in the hospital that were not present or incubating on admission; noninfectious adverse events that occurred in the hospital were all adverse events with infectious events excluded. Examples of infectious adverse events include catheter-related bloodstream infections, ventilator-associated pneumonia, and urinary tract infections, while noninfectious adverse events included events such as acute respiratory failure, acute renal failure, or oversedation. We rated adverse-event severity using a 5-point Likert scale (ie, minimal, clinically significant, serious, life-threatening, or fatal) using a severity index modified from the National Coordinating Council for Medication Error Reporting and Prevention.Reference Griffin and Resar ^{2 , 17} Adverse events that were serious (eg, organ dysfunction), life-threatening (ie, death possible within a few hours without intervention), or fatal were classified as severe events. Similarly, preventability was measured on a 4-point Likert scale (ie, definitely, probably, probably not, or definitely not preventable) as has been used previously.Reference Landrigan, Parry, Bones, Hackbarth, Goldmann and Sharek ^{3 ,} Reference Rothschild, Landrigan and Cronin ^{18 ,} Reference Sharek, Parry and Goldmann ¹⁹ Preventability was determined based on whether the event was considered preventable by current knowledge, practice standards, or technology.Reference Rothschild, Landrigan and Cronin ¹⁸ Finally, adverse events were also categorized by the primary physiological system affected (eg, cardiovascular, respiratory, renal or endocrine, hematologic, gastrointestinal, neurologic), as hospital-acquired infection, or as surgically related.Reference Landrigan, Parry, Bones, Hackbarth, Goldmann and Sharek ³ Rates of adverse events were calculated as the number of a specific adverse event type divided by total person-time at risk (ie, patient-specific hospital length of stay in days).

Statistical Analysis

Distribution of demographic variables, such as age or gender, by the use of contact precautions was analyzed using Student t tests for continuous variables and χ² tests for categorical variables. An initial multivariable, scaled-deviance Poisson regression model for noninfectious adverse events was constructed by including contact precaution status and adjusting for the matching variable of admission service (grouped as oncology, general medicine, or surgery). More than 1 adverse event during an admission was allowed per patient, but only the first eligible admission was reviewed for event occurrence; multiple admissions during the study period were not included. The model offset was the log of patient length of stay. An adjusted model for all noninfectious adverse events was constructed by adding covariates 1 at a time to the initial model in order of the greatest magnitude change produced in the log rate ratio of the contact precautions estimate until no meaningful change (<10%) occurred with the addition of new covariates. As no covariates meaningfully changed the coefficient, we included biologically plausible variables such as age and Charlson comorbidity score. Model building for preventable noninfectious adverse events followed the same approach used for the model of all noninfectious adverse events. Models of severe and preventable adverse events were constructed using noninfectious adverse events only. Due to the low number of severe noninfectious adverse events, we constructed a model for severe adverse events adjusted only for matching. For each physiological system, counts of adverse events by contact precaution status were recorded, and bivariate analysis was conducted using χ² or Fisher’s exact test, as appropriate. All analyses were conducted using SAS version 9.3 (SAS Institute, Cary, NC).

Power Calculation

Applying a previously reported frequency of adverse events among general medical inpatients (25 adverse events per 100 admissions)Reference Landrigan, Parry, Bones, Hackbarth, Goldmann and Sharek ³ to the average length of stay among the current study’s patients with whom contact precautions were not used, the expected rate of adverse events among these patients was 53.2 adverse events per 1,000 patient days. For a 2-sided test and α=0.05, we had 84% power to detect an 80% increase in adverse event rate among patients on contact precautions compared to patients not on contact precautions.