Utilization Metrics Research Articles

Costs by Site of Service for Commercially-Insured Patients with Metastatic Pancreatic Cancer Receiving Guideline-Recommended Chemotherapy: Comparing Community Oncology and Hospital Outpatient Settings.

PurposeCompare total cost of care (TCOC) for commercially-insured patients with metastatic pancreatic cancer receiving FDA-approved/NCCN Category 1 preferred regimens in community oncology or hospital outpatient settings.Patients and MethodsWe used the 2016–2019 MarketScan® and Milliman Consolidated Health Cost Guidelines Sources Database (CHSD) administrative claims data to compare utilization of healthcare services and expenditures for commercially-insured patients receiving chemotherapy in community oncology or hospital outpatient settings. We identified patients with metastatic pancreatic cancer using ICD-10 diagnosis codes in 2016–2019 MarketScan® and Milliman Consolidated Health Cost Guidelines Sources Database files. Patients were assigned to cohorts based on where they received the plurality of chemotherapy services: community oncology or hospital outpatient settings. Total cost of care (TCOC) and healthcare resource utilization metrics were calculated per line of therapy (LOT) for patients receiving similar chemotherapy regimens in each cohort, and differences between cohorts were evaluated using t-testing and chi-squared statistical methods.ResultsAlthough cohorts had similar demographics, chemotherapy regimen use, and length of therapy, the mean TCOC among all patients receiving chemotherapy in hospital outpatient settings was 41% higher compared to community oncology settings. Median TCOC was 35% higher in hospital outpatient settings than in community oncology settings. Mean admissions and readmissions per beneficiary were 7% and 16% higher, respectively, for thse treated in hospital outpatient versus community oncology settings. We observed no differences in the use of emergency department or hospice care between the cohorts.ConclusionOur study indicates that patients receiving chemotherapy at community oncology centers are associated with better or equivalent outcomes and lower costs than patients receiving the same regimen in a hospital outpatient setting.

Novel Techniques to Assess Predictive Systems and Reduce Their Alarm Burden.

Machine prediction algorithms (e.g., binary classifiers) often are adopted on the basis of claimed performance using classic metrics such as precision and recall. However, classifier performance depends heavily upon the context (workflow) in which the classifier operates. Classic metrics do not reflect the realized performance of a predictor unless certain implicit assumptions are met, and these assumptions cannot be met in many common clinical scenarios. This often results in suboptimal implementations and in disappointment when expected outcomes are not achieved. One common failure mode for classic metrics arises when multiple predictions can be made for the same event, particularly when redundant true positive predictions produce little additional value. This describes many clinical alerting systems. We explain why classic metrics cannot correctly represent predictor performance in such contexts, and introduce an improved performance assessment technique using utility functions to score predictions based on their utility in a specific workflow context. The resulting utility metrics (u-metrics) explicitly account for the effects of temporal relationships and other sources of variability in prediction utility. Compared to traditional measures, u-metrics more accurately reflect the real-world costs and benefits of a predictor operating in a realized context. The improvement can be significant. We also describe a formal approach to snoozing, a mitigation strategy in which some predictions are suppressed to improve predictor performance by reducing false positives while retaining event capture. Snoozing is especially useful for predictors that generate interruptive alarms. U-metrics correctly measure and predict the performance benefits of snoozing, whereas traditional metrics do not.

GONADOTROPIN UTILIZATION DATA CAN BE USED AS AN INDICATOR OF US ASSISTED REPRODUCTIVE TECHNOLOGY (ART) MARKET DYNAMICS

National summaries produced by the Centers for Disease Control (CDC) and Society for Assisted Reproductive Technology (SART) are instructive in highlighting ART use trends but lag the market by 2-years. Here we explore gonadotropin utilization metrics as a surrogate and more proximate marker to assess US ART utilization. Resultant forecasts may be valuable for anticipating demand and better ensuring availability of resources.

An Iterative Optimization Approach for Routing, Modulation, and Categorical Spatial Bandwidth Block Allocation to Improve Network Performance for Dynamic Traffic in Elastic Optical Networks

In this work, we have proposed an iterative optimization model for allocating spectral resources in optical networks. The proposed model gives spatial routes and spatial bandwidth allocations in optical networks with variable data-rates, modulation schemes, and optical reach adaptation. We have also proposed an algorithm which allocates continuous and contiguous block of frequency slots (FS) between transponders which forms bandwidth partitions. The primary objective of the bandwidth partition is to reduce spatial fragmentation. The integrated approach includes the routing information from using the optimization model and the categorical spectrum allocation from using the proposed algorithm. The integrated approach has been used for dynamic traffic to improve network performance in terms of bandwidth blocking, link utilization, and fragmentation metrics. It has been shown that the FS utilization (FSU) and link utilization (LU) largely increase in the proposed integrated scheme with 80% LU compared to shortest path first (SPF) routing with LU as low as 20%. Similarly, the standard deviation between FSU in the proposed scheme is approximately 5% compared to 25% in other schemes which shows that the FSU sufficiently increases in the integrated approach.

S874 Evaluation of Quality-of-Care Indicators Among Patients With Crohn’s Disease and Ulcerative Colitis in the United States: 2019–2020

Introduction: Crohn’s disease (CD) and ulcerative colitis (UC) can be difficult to manage and, due to a lack of meaningful quality measures, patient (pt) care may vary by provider. To understand where gaps in care may exist for these pts, this study assessed specific healthcare resource utilization (HRU) and medication metrics that may be potential quality of care (QOC) indicators. Methods: Using a large commercial US claims database (2019–2020), pts with CD or UC were identified. Potential QOC indicators were selected based on clinical guidelines and recommendations from measures of quality organizations and included CD or UC prevalence; gastroenterologist (GE) and IBD-related non-GE outpatient visits; IBD-related emergency department visits or hospitalizations; excessive steroid use (prednisone equivalent ≥10 mg/day for ≥60 consecutive days or a single prescription of ≥600 mg prednisone); excessive steroid users on corticosteroid (CS)-sparing therapy; excessive steroid users with central dual-energy X-ray absorptiometry (DEXA) or osteoporosis pharmacologic treatment; use of targeted immunomodulators (TIMs) and oral mesalamine (CD only); imaging assessments; and assessment of inflammatory biomarkers. National percentages of pts achieving each metric are reported. Results: In total, 41,555 CD and 52,507 UC pts were identified in 2019, resulting in a 0.3% and 0.4% prevalence, respectively (Table). Over a third of CD pts (39.8%) and almost half of UC pts (45.5%) did not visit a GE in 2019. Around 10% CD pts, and up to 6.4% of UC pts, had IBD-related ED visits or hospitalizations. 17.1% CD and 14.5% UC pts were excessive steroid users, yet < 9% CD and UC pts, received DEXA scans and/or bone treatments. A third of excessive steroid users with CD (34.5%), and over half (53.0%) of those with UC, did not receive CS-sparing therapy. The rate of TIM use was over two times higher in CD vs UC pts (CD: 44.3%; UC: 18.9%). Despite evidence that mesalamine is ineffective in CD, 18.7% of pts with CD were prescribed it. Inflammatory biomarker level testing rates were < 50% in both CD and UC. Similar outcomes were reported in 2020, with lower HRU, possibly due to COVID-19. Conclusion: This analysis of QOC indicators highlights various areas for improvement that may provide better treatment outcomes and reduce HRU for pts with CD and UC. Future research is needed to assess outcomes in pts that are not being routinely monitored. Table 1. - Proportion of CD and UC patients achieving potential QOC metrics by year Indicator 2019 2020 CDN=41,555a UCN=52,507a CDN=39,025a UCN=47,940a Diagnosis Rate 0.3% 0.4% 0.3% 0.4% GE Outpatient Visits 60.2% 54.5% 57.5% 52.9% IBD-Related Non-GE Outpatient Visit 38.5% 43.8% 41.3% 45.6% IBD-related ED Visits 11.2% 4.7% 9.9% 4.5% IBD-related Hospitalizations 10.3% 6.4% 9.1% 5.9% Excessive Steroid Useb Corticosteroid-Sparing Therapyc,d DEXA or Bone Treatmentc,e,f 17.1%65.5%8.9% 14.5%47.0%7.9% 15.8%66.9%8.0% 14.5%49.4%6.9% Targeted Immunomodulator Use 44.3% 18.9% 48.4% 22.2% Oral Mesalamineg 18.7% N/A 17.0% N/A Imagingh 59.4% 64.2% 54.0% 58.0% Inflammatory Biomarkersi 49.8% 35.8% 49.9% 36.9% a) Sample size shown is based on continuous eligibility in medical benefit. Denominator for medication use outcomes focused on patients fully enrolled in both medical and pharmacy benefit, and thus were a subset of the larger population. b) Excessive steroid use was defined as doses ≥10 mg/day prednisone equivalent for ≥60 consecutive days or a single prescription of ≥600 mg prednisone. c) In excessive steroid users only. d) Treatments included thiopurine, methotrexate, or TIMs. e) All indicators were within the respective calendar year except DEXA/bone treatments that occurred during the prior calendar year. f) Defined as prescription osteoporosis treatment, excluding vitamin and mineral supplements. g) In patients with CD only. h) Including endoscopy, CT scan, MRI, or ultrasound. i) Including fecal calprotectin and c-reactive protein. CD, Crohn’s disease; CT, computed tomography; DEXA, dual-energy X-ray absorptiometry; ED, emergency department; GE, gastroenterologist; IBD, inflammatory bowel disease; TIM, targeted immunomodulator; UC, ulcerative colitis.

Qualitative evaluation of two web-based tools to improve accessibility of evidence reports.

The Agency for Healthcare Research and Quality (AHRQ) Evidence-based Practice Center (EPC) Program produces evidence reports to assist health systems in improving patient care. To make EPC evidence reports more accessible and usable, AHRQ convened a panel of learning health system (LHS) senior leaders to develop two web-based tools-a table-based tool presenting high-level results, and a graphical display that presented detailed data-to disseminate the reports and test the tools. Researchers examined (1) the context under which interview participants reviewed and used the evidence reported by the tools, (2) their experiences using the tools, (3) the tools' influence on clinical practice, and (4) how the tools could be improved. Researchers collected and inductively analyzed qualitative data from tool implementation meetings with six LHSs and interviews with 27 LHS leaders and clinical staff who used the tools. Researchers used website utilization metrics to augment qualitative results. The tools were efficient, complementary, and useful sources of summarized evidence to promote system change, educate trainees and clinicians, inform research, and support shared decision making with patients and families. Clinical leaders appreciated the evidence review thoroughness and quality and viewed AHRQ as a trusted source of information. However, many felt the tools were not practical for bedside use because of their complex content. Participants also noted the reports had limited evidence strength and robustness. They suggested optimizing the tools for mobile device use to facilitate tool uptake and developing training resources about tool navigation and statistical content interpretation. LHSs found the tools to be useful resources for making the EPC Program reports more accessible to and usable for health system leaders. The tools have the potential to meet some, but not all, LHS evidence needs. Their value depends on reports' usefulness, which ultimately depends on the evidence quality.

Multi-Level Posterior Cervical Foraminotomy Associated With Increased Post-operative Infection Rates and Overall Re-Operation Relative to Anterior Cervical Discectomy With Fusion or Cervical Disc Arthroplasty.

Retrospective Cohort Study. Cervical radiculopathy meeting operative criteria has traditionally been managed using anterior cervical discectomy and fusion (ACDF). However, cervical disc arthroplasty (CDA) and posterior cervical foraminotomy (PCF) are also reasonable options. This study aimed to assess differences in postoperative outcomes among patients undergoing multi-level ACDF, CDA, or PCF comparing medical/surgical complications and healthcare utilization parameters. Patients who underwent multi-level ACDF, CDA, or PCF between 2012 and 2019 were identified from the American College of Surgeons National Surgical Quality Improvement Project (ACS-NSQIP) database. Patients were stratified based on procedure type and propensity score matched to resolve baseline differences. ANOVA was performed to identify differences in medical complications, surgical complications, and healthcare utilization metrics. A total of 31344 patients who underwent an eligible procedure were identified (ACDF: n = 28089, CDA: n = 1748, PCF: n = 1507), and 684 patients remained in each group following propensity score matching. Patients undergoing multi-level PCF were found to experience longer lengths of hospital stay (PCF: 1.67 ± 1.61days, ACDF: 1.50 ± 1.32days, CDA: 1.27 ± 1.05days, P < .001), higher rates of reoperation (PCF: 3.2%, ACDF: 1.0%, CDA: .4%, P = .020), superficial infection (PCF: 1.3%, ACDF: .3%, CDA: .1%, P = .008) and deep infection (PCF: 1.2%, ACDF: 0%, CDA: 0%, P < .001). There were no outcome differences between multi-level ACDF and CDA. Patients undergoing multi-level PCF were at increased risk for longer hospital stay, re-operation, and infection relative to those undergoing ACDF and CDA. Future research should aim to uncover the precise mechanisms underlying these complications, as well as analyze long term outcomes. III.

Prescribing Narcotics for Pain: Reconsider the Fifth Vital Sign.

Prescribing Narcotics for Pain: Reconsider the Fifth Vital Sign.

Demographic and clinical profile of patients treated with proximal femoral nails – a 10-year analysis of more than 40,000 Cases

BackgroundHip fractures are common in elderly populations and can be life threatening. Changes in healthcare delivery and outcomes for patients with hip fracture treated with intramedullary nails are not well characterized. The objectives of our study were: 1) the characterization of patients treated with the Trochanteric Fixation Nail -Advanced™(TFNA) Proximal Femoral Nailing System or comparable nails (index) and estimate 12-month all-cause readmissions (ACR) and reoperations following index; and 2) the evaluation of 10-year healthcare utilization (HCU) trends for treatment of femoral fractures with femoral nails.MethodsThis is a retrospective database analysis using the Premier hospital database. All adults with femoral fracture treated with an intramedullary nail, from 2010 to Q3 2019, in the inpatient setting, were identified. Exclusion criteria included patients with bilateral hip surgery and presence of breakage at time of initial surgery. The primary outcome was ACR and reoperation, the secondary outcomes were healthcare utilization metrics. Variables included demographics, comorbidities (Elixhauser Index (EI)), surgical intervention variables and hospital characteristics.ResultsForty-one thousand one hundred four patients were included in the study, of which 14,069 TFNA patients, with average age 77.9 (Standard deviation (SD): 12.0), more than 60% with 3 or more comorbidities (more than 64% for TFNA), 40% with severe or extreme disease severity and one third with severe or extreme risk for mortality. ACR reached 60.1% (95% confidence interval (CI): 59.6%-60.5%) – for TFNA: 60.0% (95%CI: 59.2%-60.8%). The reoperation rate was 4.0% (95%CI: 3.8%-4.2%) – for TFNA: 3.8% (95%CI: 3.5%-4.1%). Length of stay (LOS) averaged 5.8 days (SD: 4.8), and 12-month hip reoperation was 4.0% (3.8%-4.2%), in TFNA cohort: 3.8% (3.5%-4.1%). From 2010 to 2019: the percentage patients operated within 48 h of admission significantly increased, from 75.2% (95%CI: 74.3%-76.1%) to 84.3% (95%CI: 83.9%-84.6%); LOS significantly decreased, from 6.2 (95%CI: 6.0–6.4) to 5.6 (95%CI: 5.5–5.7) days; discharge to skilled nursing facilities (SNF) increased from 56.0% (95%CI: 54.8%-57.2%) to 61.5% (95%CI: 60.8%-62.2%); ACR rates decreased but reoperation rates remained constant.ConclusionsACR and reoperation rates were similar across device types and averaged 60.1% and 4.0%, respectively. Ten-year analyses showed reductions in hospital HCU and greater reliance on SNF.

Incidence of repeat procedures and healthcare utilization following surgery, radiosurgery, and percutaneous procedures in elderly patients with trigeminal neuralgia.

Management of trigeminal neuralgia (TN) in elderly patients poses significant challenges. The impact of different treatment modalities (surgery, radiosurgery [RS], and percutaneous techniques [PTs]) on healthcare utilization is not well defined in the management of TN in elderly patients. The aim of this study was to compare the long-term healthcare utilization metrics of different interventions in the management of elderly patients with TN. The MarketScan database was queried using the International Classification of Diseases, Ninth Revision and Current Procedural Terminology, from 2000 to 2016. TN patients ≥ 65 years of age managed using surgery, RS, and PTs with at least 5 years of follow-up after the index procedure were included. Outcomes analyzed were hospital admissions, outpatient services, and medication refills. Of 993 patients, 43% (n = 430) underwent RS, 44% (n = 432) had PTs, and only 13% (n = 131) underwent surgery for TN. Overall, the median age of patients was 74 years old, 64% were females, 90% had Medicare insurance, and 17% had an Elixhauser index ≥ 3. Patients in the surgery group were younger (median age 71 years) with a higher comorbidity index (≥ 3; 24%) compared with patients undergoing RS and PTs (13% and 17%, respectively). At 1, 2, and 5 years after the index procedure, 41%, 48%, and 57% of patients in the PT cohort underwent any repeat procedure compared with 11%, 18%, and 29% for the RS cohort, and 6%, 9%, and 11% for the surgical cohort, respectively. Also, patients in the PT cohort incurred 1.8, 1.9, and 2.0 times the combined payment at 1, 2, and 5 years, respectively, compared with the surgery cohort. Similarly, patients who underwent RS for TN incurred 1.4, 1.5, and 1.5 times the combined payment at 1, 2, and 5 years, respectively, compared with the surgery cohort. At 5 years after the index procedure, combined payments for the PT cohort were $79,753 (IQR $46,013, $144,064) compared with $61,016 (IQR $27,114, $117,097) for the RS cohort and $41,074 (IQR $25,392, $87,952) for the surgery cohort (p < 0.0001). PTs followed by RS were the common procedures used in the majority of elderly patients with TN. However, surgery for TN resulted in durable control with the least need for reoperations up to 5 years after the index procedure, followed by RS and PTs. PTs for TN resulted in the highest utilization of healthcare resources and need for reoperations at all time points. These findings should be considered in clinical decision-making when selecting appropriate treatment modalities in elderly patients with TN.

Predictors of Extreme Hospital Length of Stay After Traumatic Brain Injury

Hospital length of stay (HLOS) after traumatic brain injury (TBI) is an important metric of injury severity, resource utilization, and access to post-acute care services. Risk factors for protracted HLOS after TBI require further characterization. Data regarding adult inpatients admitted to a single U.S. level 1 trauma center with a diagnosis of acute TBI between August 1, 2019, and April 1, 2022, were extracted from the electronic health record. Patients with extreme HLOS (XHLOS, >99th percentile of institutional TBI HLOS) were compared with those without XHLOS. Socioeconomic status (SES), clinical/injury factors, and discharge disposition were analyzed. In 1638 patients, the median HLOS was 3 days (interquartile range [IQR]: 2-8 days). XHLOS threshold was >70 days (N= 18; range: 72-146 days). XHLOS was associated with younger age (XHLOS/non-XHLOS: 50.4/59.6 years; P= 0.042) and greater proportions with severe TBI (55.6%/11.4%; P < 0.001), low SES (72.2%/31.4%; P < 0.001), and Medicaid insurance (77.8%/30.1%; P < 0.001). XHLOS patients were more likely to die in hospital (22.2%/8.1%) and discharge to post-acute facility (77.8%/16.3%; P < 0.001). No XHLOS patients were discharged to home. In XHLOS patients alive at discharge, medical stability was documented at median 39 days (IQR: 28-58 days) and were hospitalized for another 56 days (IQR: 26.5-78.5 days). XHLOS patients were more likely to have severe injuries, low SES, and Medicaid. XHLOS is associated with in-hospital mortality and need for post-acute placement. XHLOS patients often demonstrated medical stability long before placement, underscoring complex relationships between SES, health insurance, and outcome. These findings have important implications for quality improvement and resource utilization at acute care hospitals and await validation from larger trials.

#Ophthalmology: Social Media Utilization and Impact in Ophthalmology Journals, Professional Societies, and Eye Health Organizations.

AimTo evaluate and quantify social media presence of ophthalmology peer-reviewed journals, professional societies, and eye health organizations, and to determine if there is a correlation between social media utilization and Twitter engagement metrics or journal impact measures.MethodsWe searched for online profiles of 100 ophthalmology peer-reviewed journals, 333 professional ophthalmology societies, and 40 eye health organizations on Facebook, Twitter, and Instagram. Impact was quantified by recording the number of “likes” on Facebook and number of followers on Twitter and Instagram. We also used Twitonomy software to obtain advanced Twitter metrics for all journal accounts from 2018 to 2021, and compared to journal impact measured by SCImago Journal Rank (SJR) score, the h-index, and impact factor.ResultsEye health organizations averaged significantly greater Facebooks “likes” and Twitter followers than both peer-reviewed journals and professional societies (p < 0.0001). Of 100 journals studied, 30% were active on Twitter, 25% on Facebook, and 6% on Instagram. Slightly more than half of all journal-affiliated social media accounts were accessible from the journal website. Among journals with active Twitter accounts, total followers, total tweets, average retweets, and average “favorites” were all significantly positively correlated with academic metrics such as the SJR, h-index, and/or impact factor.ConclusionGreater social media engagement is associated with higher ophthalmology journal impact metrics; however, ophthalmology journals and professional societies lag behind eye health organizations in social media engagement. Although unable to demonstrate causality, social media may be an underutilized visibility, communication, and dissemination tool.

Enabling Intelligent IoTs for Histopathology Image Analysis Using Convolutional Neural Networks.

Medical imaging is an essential data source that has been leveraged worldwide in healthcare systems. In pathology, histopathology images are used for cancer diagnosis, whereas these images are very complex and their analyses by pathologists require large amounts of time and effort. On the other hand, although convolutional neural networks (CNNs) have produced near-human results in image processing tasks, their processing time is becoming longer and they need higher computational power. In this paper, we implement a quantized ResNet model on two histopathology image datasets to optimize the inference power consumption. We analyze classification accuracy, energy estimation, and hardware utilization metrics to evaluate our method. First, the original RGB-colored images are utilized for the training phase, and then compression methods such as channel reduction and sparsity are applied. Our results show an accuracy increase of 6% from RGB on 32-bit (baseline) to the optimized representation of sparsity on RGB with a lower bit-width, i.e., <8:8>. For energy estimation on the used CNN model, we found that the energy used in RGB color mode with 32-bit is considerably higher than the other lower bit-width and compressed color modes. Moreover, we show that lower bit-width implementations yield higher resource utilization and a lower memory bottleneck ratio. This work is suitable for inference on energy-limited devices, which are increasingly being used in the Internet of Things (IoT) systems that facilitate healthcare systems.

P36. Multilevel posterior cervical foraminotomy associated with increased perioperative infection rates relative to anterior cervical discectomy with fusion and cervical disc arthroplasty

P36. Multilevel posterior cervical foraminotomy associated with increased perioperative infection rates relative to anterior cervical discectomy with fusion and cervical disc arthroplasty

129. National trends in the utilization of lumbar disc replacement for lumbar degenerative disc disease over a 10-year period, 2010 to 2019

<h3>BACKGROUND CONTEXT</h3> In the treatment of lumbar degenerative disc disease (DDD) that has failed conservative measures, lumbar fusion (LF) remains the gold standard. Approved in 2004, lumbar disc replacement (LDR) is designed to preserve motion. Initial studies showed LDR was used in 2.7% of surgeries for DDD and spondylosis. <h3>PURPOSE</h3> We aim to analyze the trends, complications and costs associated with LDR compared to LF using a large national database. <h3>STUDY DESIGN/SETTING</h3> Restrospective, administrative database. <h3>PATIENT SAMPLE</h3> National Inpatient Sample, 2010-2019. <h3>OUTCOME MEASURES</h3> Cost, length of stay, discharge destination, index hospitalization complications. <h3>METHODS</h3> The National Inpatient Sample (NIS) was queried from 2010 to 2019 to identify patients undergoing single and double- level LF and LDR with a diagnosis of lumbar DDD using ICD-9 and ICD-10 diagnostic and procedure codes. Revision procedures were excluded. Propensity score matching (PSM) with a ratio of 2:1 was performed to compare health care utilization metrics and index hospital complications for patients undergoing single-level LF or LDR. <h3>RESULTS</h3> A total of 1,129,121 LF cases (99.3%) and 8,049 LDR cases (0.7%) we identified during the study period, with 364,637 (32.3%) and 712 (8.8%) comprising two-level procedures, respectively. 1,712 LDRs were performed in 2010 (1.27% of all), decreasing to 565 in 2013 (0.52%), and increased slightly to 870 in 2019 (0.74%). For single-level procedures, the proportion of LDR cases decreases from 1.41% in 2010 to 0.66% in 2013 (p <0.001), but then increased gradually to 1.17% in 2019 (p <0.001 for trend). The proportion of two-level LF and LDR procedures increased significantly over the study period from 15.7% in 2010 to 45.5% in 2019 (p <0.001 for trend). LDR patients were significantly more likely to be younger (mean age 41.2 vs 57.1, p <0.001), healthier (mean ECI 0.88 vs 1.80, p <0.001) and male (57.5% vs 47.1%, p <0.001). LDR patients were also more likely to have procedures paid for by private insurance (55.8% vs 43.7%, p <0.001) and reside in higher income areas (top quartile 30.0% vs 23.2%, p <0.001). On PSM analysis, a total of 1,343 single-level LDR patient records were matched to 2,686 single-level LF patient records, and 141 two-level LDR patient records were matched to 282 two-level LF patient records. Single-level LDR patients had significantly shorter LOS (2.45 vs 3.10 days, p <0.001), lower likelihood of discharge to facility (1.7% vs 3.4%, p=0.003), and lower costs overall ($27,714 vs $32,242, p <0.001) compared to matched LF patients. Single-level LDR patients had significantly lower rates of any complication (7.0% vs 13.2%, p <0.001), blood transfusion (3.1% and 8.1%, p <0.001) and neurologic complication (3.0% vs 4.8%, p=0.006). There were no differences with regard to VTE, pneumonia, renal, would or pulmonary complications. With regard to two-level procedures, there were no differences with regard to LOS and discharge to facility (p=0.38 and 0.21, respectively), while overall costs were significantly higher for LDR patients compared to LF ($53,270 to $44,721, p=0.005). There were no differences in rates of index hospitalization complications for two-level procedures. <h3>CONCLUSIONS</h3> The utilization of LDR procedures decreased from 2010-2017 but began to increase again in 2018 and 2019. After adjusting for demographic, medical and hospital variables, single-level LDR was associated with reduced costs and LOS, and lower rates of blood transfusion compared to single-level fusion in patients with lumbar DDD. <h3>FDA DEVICE/DRUG STATUS</h3> This abstract does not discuss or include any applicable devices or drugs.

Using Saturated Count Models for User-Friendly Synthesis of Large Confidential Administrative Databases

AbstractOver the past three decades, synthetic data methods for statistical disclosure control have continually evolved, but mainly within the domain of survey data sets. There are certain characteristics of administrative databases, such as their size, which present challenges from a synthesis perspective and require special attention. This paper, through the fitting of saturated count models, presents a synthesis method that is suitable for administrative databases. It is tuned by two parameters, σ and α. The method allows large categorical data sets to be synthesized quickly and allows risk and utility metrics to be satisfied a priori, that is, prior to synthetic data generation. The paper explores how the flexibility afforded by two-parameter count models (the negative binomial and Poisson-inverse Gaussian) can be utilised to protect respondents'—especially uniques'—privacy in synthetic data. Finally, an empirical example is carried out through the synthesis of a database which can be viewed as a good substitute to the English School Census.

AN ADVERSARIAL ROBUSTNESS ANALYSIS FRAMEWORK FOR CLASSIFICATION MODEL

In the ML/DL algorithms developed by researchers/developers, the only constraint faced by them is the robustness of the model. The model may be experimented with certain data sets but for the model to get tested with all the permutation and combination of data sets under different conditions is a challenge. For the model to give efficient and accurate results under all circumstances, the users are demanding a more robust model that has been analyzed thoroughly. We aim to create an algorithm that is unique, efficient and solves a real-world problem and build an Adversarial Robustness analysis framework that works on reinforcement learning, applies utility metrics of attacks and defense to give a comprehensive robustness report

Multi-Stage Hybrid Federated Learning Over Large-Scale D2D-Enabled Fog Networks

Federated learning has generated significant interest, with nearly all works focused on a "star" topology where nodes/devices are each connected to a central server. We migrate away from this architecture and extend it through the network dimension to the case where there are multiple layers of nodes between the end devices and the server. Specifically, we develop multi-stage hybrid federated learning (MH-FL), a hybrid of intra- and inter-layer model learning that considers the network as a multi-layer cluster-based structure. MH-FL considers the topology structures among the nodes in the clusters, including local networks formed via device-to-device (D2D) communications, and presumes a semi-decentralized architecture for federated learning. It orchestrates the devices at different network layers in a collaborative/cooperative manner (i.e., using D2D interactions) to form local consensus on the model parameters and combines it with multi-stage parameter relaying between layers of the tree-shaped hierarchy. We derive the upper bound of convergence for MH-FL with respect to parameters of the network topology (e.g., the spectral radius) and the learning algorithm (e.g., the number of D2D rounds in different clusters). We obtain a set of policies for the D2D rounds at different clusters to guarantee either a finite optimality gap or convergence to the global optimum. We then develop a distributed control algorithm for MH-FL to tune the D2D rounds in each cluster over time to meet specific convergence criteria. Our experiments on real-world datasets verify our analytical results and demonstrate the advantages of MH-FL in terms of resource utilization metrics.

The Impact of the COVID-19 Pandemic on Health Care Utilization Among Insured Individuals With Common Chronic Conditions.

Objective:The COVID pandemic has had a significant impact on the US health care system. Our primary objective was to understand the impact of the COVID pandemic on non–COVID-related health care utilization among insured individuals with chronic conditions. Our secondary objective was to examine the differential impact by individual characteristics.Main Data Source:Medical and pharmacy claims data for individuals enrolled in a large insurer across the United States.Research Design:A retrospective and repeated cross-sectional study. Overall and condition-specific health care utilization and cost metrics in (1) March 1 to June 15 and (2) June 16 to September 30, 2020 were compared with the same months during 2016–2019.Subjects:Members of all ages with a diagnosis of diabetes, cardiovascular disease, or chronic kidney disease with commercial or Medicare Advantage insurance.Results:Most non–COVID-related health care utilization decreased drastically on March 1 to June 15, 2020 [odds ratio (OR) range across condition-specific tests: 0.55–0.69; incidence rate ratio (IRR) range for hospitalization/emergency department (ED) visit/outpatient visit: 0.65–0.77] but returned to closer to pre-COVID levels by June 16 to September 30, 2020 [OR range across condition-specific tests: 0.93–1.08; IRR range for hospitalization/ED visit/outpatient visit: 0.77–0.97]. Our study found an enormous increase in telehealth use on March 1 to June 15, 2020 (90–170 times prepandemic levels). A differential impact was observed by age, sex, region of residence, and insurance type.Implications:Further investigation is needed to assess the impact of these changes in health care utilization on long-term health outcomes.

Privacy-utility trades in crowdsourced signal map obfuscation

Cellular providers and data aggregating companies crowdsource cellular signal strength measurements from user devices to generate signal maps, which can be used to improve network performance. Recognizing that this data collection may be at odds with growing awareness of privacy concerns, we consider obfuscating such data before the data leaves the mobile device. The goal is to increase privacy such that it is difficult to recover sensitive features from the obfuscated data (e.g. user ids and user whereabouts), while still allowing network providers to use the data for improving network services (i.e. create accurate signal maps). To examine this privacy-utility tradeoff, we identify privacy and utility metrics and threat models suited to signal strength measurements. We then obfuscate the measurements using several preeminent techniques, spanning differential privacy, generative adversarial privacy, and information-theoretic privacy techniques, in order to benchmark a variety of promising obfuscation approaches and provide guidance to real-world engineers who are tasked to build signal maps that protect privacy without hurting utility. Our evaluation results, based on multiple, diverse, real-world signal map datasets, demonstrate the feasibility of concurrently achieving adequate privacy and utility, with obfuscation strategies which use the structure and intended use of datasets in their design, and target average-case, rather than worst-case, guarantees.