A Simulated Prospective Evaluation of a Deep Learning Model for Real-Time Prediction of Clinical Deterioration Among Ward Patients.

OBJECTIVES: The National Early Warning Score, Modified Early Warning Score, and quick Sepsis-related Organ Failure Assessment can predict clinical deterioration. These scores exhibit only moderate performance and are often evaluated using aggregated measures over time. A simulated prospective validation strategy that assesses multiple predictions per patient-day would provide the best pragmatic evaluation. We developed a deep recurrent neural network deterioration model and conducted a simulated prospective evaluation. DESIGN: Retrospective cohort study. SETTING: Four hospitals in Pennsylvania. PATIENTS: Inpatient adults discharged between July 1, 2017, and June 30, 2019. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: We trained a deep recurrent neural network and logistic regression model using data from electronic health records to predict hourly the 24-hour composite outcome of transfer to ICU or death. We analyzed 146,446 hospitalizations with 16.75 million patient-hours. The hourly event rate was 1.6% (12,842 transfers or deaths, corresponding to 260,295 patient-hours within the predictive horizon). On a hold-out dataset, the deep recurrent neural network achieved an area under the precision-recall curve of 0.042 (95% CI, 0.04-0.043), comparable with logistic regression model (0.043; 95% CI 0.041 to 0.045), and outperformed National Early Warning Score (0.034; 95% CI, 0.032-0.035), Modified Early Warning Score (0.028; 95% CI, 0.027- 0.03), and quick Sepsis-related Organ Failure Assessment (0.021; 95% CI, 0.021-0.022). For a fixed sensitivity of 50%, the deep recurrent neural network achieved a positive predictive value of 3.4% (95% CI, 3.4-3.5) and outperformed logistic regression model (3.1%; 95% CI 3.1-3.2), National Early Warning Score (2.0%; 95% CI, 2.0-2.0), Modified Early Warning Score (1.5%; 95% CI, 1.5-1.5), and quick Sepsis-related Organ Failure Assessment (1.5%; 95% CI, 1.5-1.5). CONCLUSIONS: Commonly used early warning scores for clinical decompensation, along with a logistic regression model and a deep recurrent neural network model, show very poor performance characteristics when assessed using a simulated prospective validation. None of these models may be suitable for real-time deployment.

Analysis of perioperative factors associated with increased cost following abdominal wall reconstruction (AWR).

BACKGROUND: Ventral hernias are a common, challenging, and expensive problem for both the general and reconstructive surgeons; therefore, the aim of this study is to critically assess perioperative factors related to cost in abdominal wall reconstructions (AWR). METHODS: A retrospective review of AWR patients from 2007 and 2012 was performed. Analysis of perioperative factors associated with total cost of reconstruction was performed. Linear regression analyses were used to assess independent predictors of total cost. RESULTS: 134 consecutive AWR performed by a single surgeon over a 5-year period at an academic teaching center were included. The average total cost of AWR was $61,251 ± 55,624. Linear regression analysis demonstrated that diabetes (P = 0.026), increased American Society of Anesthesiologists score (P = 0.002), preoperative anemia (P = 0.001), and hernias derived from trauma (P = 0.015) were independently associated with added cost in AWR when controlling for confounding variables. In addition, patients requiring intra-abdominal procedures (P = 0.012) and those receiving an AWR using Acellular Dermal Matrix (P = 0.015) accrued significantly greater cost. Interestingly, preoperative placement of an epidural (P = 0.011) was independently associated with significant cost savings and reduced medical morbidity. Major surgical complications (P < 0.001) and length of stay (P < 0.001) were independently associated with increased cost following AWR. CONCLUSION: We present a critical assessment of cost in AWR at a major academic teaching hospital and quantify the impact of reconstruction in the setting of medical morbidities and reconstructive complexities. The data from this study can be used to adjust reimbursement schemes and to critically assess the cost-benefit of performing AWR.

Early hospital readmission in the trauma population: are the risk factors different?

INTRODUCTION: Hospital readmission rates will soon impact Medicare reimbursements. While risk factors for readmission have been described for medical and elective surgical patients, little is known about their predictive value specifically in trauma patients. PATIENTS AND METHODS: We retrospectively identified all admissions after trauma resuscitation to our urban level 1 trauma centre from 1/1/2004 to 8/31/2010. All patients discharged alive were included. Data collected included demographics, Injury Severity Score (ISS), and length of stay (LOS). We analyzed these index admissions for the development of complications that have previously been shown to be associated with readmission. Readmissions that occurred within 30 days of index admission were identified. Univariable and multivariable analyses were performed. p<0.05 was considered significant. RESULTS: We identified 10,306 index admissions, with 447 (4.3%) early (within 30 days) readmissions. Mean ISS was 11.1 (SD 10.4). On multivariable analysis, African-American race (OR 1.3, p=0.009), pre-existing chronic obstructive pulmonary disease (COPD) (OR 1.5, p=0.02), and diabetes mellitus (OR 1.8, p<0.001) were associated with readmission, along with higher ISS (OR 1.01, p<0.001), ICU admission (OR 2.1, p<0.001), and increased LOS (OR 1.01, p<0.001). Among many in-hospital complications examined, only the development of surgical site infection (SSI) (OR 1.9, p=0.02) was associated with increased risk of readmission. CONCLUSIONS: Trauma patients have a low risk of readmission. In contrast to elective surgical patients, the only modifiable risk factor for readmission in our trauma population was SSI. Other risk factors may present clinicians with opportunities for targeted interventions, such as proactive follow up or early phone contact. With future changes to health care policy, clinicians may have even greater motivation to prevent readmission.

Breast reconstruction modality outcome study: a comparison of expander/implants and free flaps in select patients.

BACKGROUND: Choosing a breast reconstructive modality after mastectomy is a critical step involving complex decisions. The authors provide outcomes data comparing two common reconstructive modalities to assist patients and surgeons in preoperative counseling and discussions. METHODS: A prospectively maintained database was queried identifying select patients undergoing expander/implant and abdominally based free flaps for breast reconstruction between 2005 and 2008. Variables evaluated included comorbidities, operations, time to reconstruction, complications, overall outcome, clinic visits, revisions, and costs. RESULTS: One hundred forty-two patients received free flaps and 60 received expander/implants. Expander/implant patients required more procedures (p < 0.001) but had shorter overall hospital lengths of stay (p < 0.001). The two cohorts experienced a similar rate of revision (p = 0.17). Free flap patients elected to undergo nipple-areola reconstruction more frequently (p = 0.01) and were able to sooner (p < 0.0001). Patients undergoing expander/implant reconstruction had a higher rate of failure (7.3 versus 1.3 percent, p = 0.008). Free flap patients achieved a stable reconstruction significantly faster (p = 0.0005), with fewer visits (p = 0.02). Cost analysis demonstrated total cost trended toward significantly lower in the free flap cohort (p = 0.15). Reconstructive modality was the only independent factor associated with time to stable reconstruction and reconstructive failure (p < 0.001 and p = 0.05, respectively). CONCLUSIONS: The authors' analysis revealed that free flap reconstructions required fewer procedures, had lower rates of complications and failures, had fewer clinic visits, and achieved a final, complete reconstruction faster than expander/implant reconstructions. Although autologous reconstruction is still not ideal for every patient, these findings can be used to enhance preoperative discussions when choosing a reconstructive modality. CLINICAL QUESTION/LEVEL OF EVIDENCE: : Therapeutic, III.

The influence of unit-based nurse practitioners on hospital outcomes and readmission rates for patients with trauma.

BACKGROUND: With the increased restrictions on resident work hours, hospitals increasingly are relying on advance practice nurses and physician assistants to help meet the patient care demand. We have created a workflow model wherein unit-based nurse practitioners (UBNPs) provide the minute-to-minute care for patients with trauma in one specific unit in our hospital, with supervision by the attending surgeons. Patients with trauma may also be admitted to other units, where the care model is a traditional resident-run (RR) service, again with supervision by the attending staff. Our aim was to determine if there were differences between the care provided by UBNPs and residents. METHODS: We queried our trauma database for all patients admitted to our urban, academic, Level I trauma center from January 1, 2007, to August 31, 2010. Patients discharged alive from the trauma service were identified and cross-referenced with an administrative database to collect demographics, injury characteristics, comorbidities, complications, and discharge information. Patients cared for by the UBNPs were compared with those cared for by the RR service. χ², Fisher's exact, and Student's t tests were used to determine significance. Significant factors were then tested with a multivariate linear regression analysis. p < 0.05 was considered significant. RESULTS: During the study period, 3,859 patients were discharged alive from the trauma service, 2,759 (71.5%) from the UBNPs service, and 1,100 (28.5%) from the RR service. Demographic data and mean Injury Severity Score (11.6 vs. 11.1, p = 0.24) were similar for the two groups, although mean abdominal Abbreviated Injury Score was higher for the UBNP group (0.6 vs. 0.5, p = 0.02). UBNP patients were more likely to be diagnosed with deep venous thrombosis (4% vs. 2.5%, p = 0.02) and were more likely to be discharged to home (67% vs. 60%, p = 0.002). Mean (SD) length of stay for UBNP patients was 6.5 (8.8) days compared with 7 (10.8) days for RR patients, although this difference did not reach statistical significance ( p = 0.17). The 30-day hospital readmission rates were similar for both groups (4.0% vs. 4.4%, p = 0.63). CONCLUSION: Care provided by UBNPs is equivalent to that provided by residents. With the restriction on resident work hours and greater reliance on nurse practitioners, patient care does not suffer. Moreover, a difference of 0.5 days in mean length of stay for the UBNP patients equates with more than 1,300 fewer patient care days. This difference, although not statistically significant, may be clinically relevant to physicians and administrators and may offset the cost of hiring UBNPs to help meet the patient care demand.

Prophylactic muscle flaps in vascular surgery: the Penn Groin Assessment Scale.

BACKGROUND: High-risk patients undergoing vascular procedures through inguinal incisions experience significant benefit from undergoing prophylactic muscle flaps. The authors present a risk assessment tool, an evidence-based algorithm for using prophylactic muscle flaps, and a comprehensive cost analysis. METHODS: Data derived from the authors' previously performed retrospective cohort study of patients undergoing open vascular groin surgery were used to construct an evidence-based risk assessment tool. A multivariate regression analysis identified significant, independent predictors of complications, which were used to construct a scoring system. An institutional cost analysis and preoperative assessment algorithm were derived based on patient risk stratification and statistical analyses. RESULTS: Sixty-eight prophylactic flaps in 53 patients were compared with 195 open femoral access procedures without flaps in 178 patients. Multivariate regression demonstrated that obesity, smoking, reoperation, and prosthetic graft reconstruction are significant predictors of complications. A weighted risk factor score (0 to 7) was devised: obesity, for a value of 1; smoking, 2; reoperation for open groin surgery, 2; and prosthetic graft material, 2. Patients with higher scores had significantly more complications, infections, and more frequently required secondary salvage flap procedures. Using study data, the authors constructed an algorithm to guide preoperative groin assessment and use of prophylactic muscle flaps. CONCLUSIONS: The authors provide an assessment tool, called the Penn Groin Assessment Scale, that accurately predicts groin complications. They also describe a simple algorithm to assess for prophylactic muscle coverage. Their results suggest that patients with two or more risk factors will benefit from prophylactic muscle flaps. CLINICAL QUESTION/LEVEL OF EVIDENCE: Risk, II.

The surgical revolving door: risk factors for hospital readmission.

BACKGROUND: Unplanned hospital readmissions increase healthcare costs and patient morbidity. We sought to identify risk factors associated with early readmission in surgical patients. MATERIALS AND METHODS: All admissions from a mixed surgical unit during 2009 were retrospectively reviewed and unplanned readmissions within 30 d of discharge were identified. Demographic data, length of stay, pre-existing diagnoses, and complications during the index admission were evaluated. T-tests and Fisher exact tests were used to examine the relationship of independent variables with readmission. Univariate and multivariate regression analysis were performed. RESULTS: A total of 1808 index admissions occurred during the study period. In all, 51 (3%) patients were readmitted within 30 d of discharge. The majority of readmissions (53%) were for infectious reasons. On univariate analyses, DVT (P = 0.004) and acute renal failure (P = 0.002) were associated with increased risk of readmission. Readmitted patients were also more likely to have public insurance (63% versus 37%, P = 0.03) and have a longer stay in the hospital (8 d, range 4-14 d versus 3 d, range 2-7 d, P = 0.001). Initial admission after trauma evaluation was associated with a decreased risk of readmission (OR 0.374, P = 0.004). Other demographic variables and pre-existing conditions were not associated with increased readmission. On multivariate logistic regression only DVT (P = 0.039) and LOS (P = 0.014) remained significant. CONCLUSIONS: Increased LOS and the development of a DVT are risk factors for early unplanned hospital readmission. Admission following trauma is associated with a decreased risk of readmission, possibly due to proactive multidisciplinary discharge planning and geographically-based nurse practitioner involvement.

Improving resource utilization in a teaching hospital: development of a nonteaching service for chest pain admissions.

PURPOSE: Exclusion of acute coronary syndrome frequently prompts a brief hospital admission for a large proportion of patients presenting to the emergency department with chest pain. At hospitals with residency programs, the volume of such patients creates pressures on these programs because of the limit on the number of patients a resident can accept in a given period. These restrictions have been instituted by the Accreditation Council for Graduate Medical Education (ACGME). The authors hypothesized that a nonteaching service designed to identify and admit low-risk chest pain patients should reduce those pressures. METHOD: A hospitalist-directed nonteaching service (NTS) was created to admit low-risk chest pain patients at the Hospital of the University of Pennsylvania. Patients' admission service was based upon the thrombolysis in myocardial infarction (TIMI) risk score. From September 2003 to June 2004, patients (n = 113) with scores of 0 or 1 (showing low risk) were admitted to the NTS. Simultaneously, a similar group of low-risk chest pain patients (n = 205) were admitted to a traditional internal medicine resident-based service (RBS). RESULTS: The NTS patients had a lower median length of stay (23 hours versus 33 hours; p < .0001) and lower median hospital charges ($8,545 versus $14,150; p < .0001) when compared with the low-risk patients on the RBS. CONCLUSIONS: The development of an NTS for chest pain admissions can assist residency programs in their efforts to meet the ACGME program requirements. The TIMI risk score can be used as a tool to assist in the identification of low-risk chest pain patients.