key: cord-0700905-ddw6ir08 authors: Kip, K. E.; Snyder, G.; Yealy, D. M.; Mellors, J. W.; Minnier, T.; Donahoe, M. P.; McKibben, J.; Collins, K.; Marroquin, O. C. title: Temporal Changes in Clinical Practice with COVID-19 Hospitalized Patients: Potential Explanations for Better In-Hospital Outcomes date: 2020-09-29 journal: nan DOI: 10.1101/2020.09.29.20203802 sha: 4993372e8ecc2593e9a3c213d8115c274681c349 doc_id: 700905 cord_uid: ddw6ir08 Background/Aims: We reviewed demographic and clinical profiles, along with measures of hospital-based clinical practice to identify temporal changes in clinical practice that may have affected in-hospital outcomes of patients with COVID-19. Methods: Data consisted of sociodemographic and clinical data captured in University of Pittsburgh Medical Center (UPMC) electronic medical record (EMR) systems, linked by common variables (deidentified). The analysis population included hospitalized patients (across 21 hospitals) with a primary diagnosis of COVID-19 infection during the period March 14-August 31, 2020. The primary outcome was a composite of in-hospital mechanical ventilation/mortality. We compared temporal trends in patient characteristics, clinical practice, and hospital outcomes using 4 time-defined epochs for calendar year 2020: March 14-March 31 (epoch 1); April 1-May 15, (epoch 2), May 16-June 28 (epoch 3); and June 29-August 31 (epoch 4). We report unadjusted survival estimates, followed by propensity score analyses to adjust for differences in patient characteristics, to compare in-hospital outcomes of epoch 4 patients (recently treated) to epoch 1-3 patients (earlier treated). Results: Mean number of hospital admissions was 9.9 per day during epoch 4, which was ~2- to 3-fold higher than the earlier epochs. Presenting characteristics of the 1,076 COVID-19 hospitalized patients were similar across the 4 epochs, including mean age. The crude rate of mechanical ventilation/mortality was lower in epoch 4 patients (17%) than in epoch 1-3 patients (23% to 35%). When censoring for incomplete patient follow-up, the rate of mechanical ventilation/mortality was lower in epoch 4 patients (p<0.0001), as was the individual component of mechanical ventilation (p=0.0002) and mortality (p=0.02). In propensity score adjusted analyses, the in-hospital relative risk (RR) of mechanical ventilation/mortality was lower in epoch 4 patients (RR=0.67, 95% CI: 0.48, 0.93). For the outcome being discharged alive within 3, 5, or 7 days of admission, adjusted odds ranged from 1.6- to 1.7-fold higher among epoch 4 patients compared to earlier treated patients. The better outcomes in epoch 4 patients were principally observed in patients under the age of 75 years. Patient level dexamethasone use was 55.6% in epoch 4 compared to 15% or less of patients in the earlier epochs. Most patients across epochs received anticoagulation drugs (principally heparin). Overall steroid (81.7% vs. 54.3%, p<0.0001) and anticoagulation use (90.4% vs. 80.7%, p=0.0001) was more frequent on the day or day after hospitalization in epoch 4 patients compared to earlier treated patients. Conclusions: In our large system, recently treated hospitalized COVID-19 patients had lower rates of in-hospital mechanical ventilation/mortality and shorter length of hospital stay. Alongside of this was a change to early initiation of glucocorticoid therapy and anticoagulation. The extent to which the improvement in patient outcomes was related to changes in clinical practice remains to be established. Conclusions: In our large system, recently treated hospitalized COVID-19 patients had lower rates of in-hospital mechanical ventilation/mortality and shorter length of hospital stay. Alongside of this was a change to early initiation of glucocorticoid therapy and anticoagulation. The extent to which the improvement in patient outcomes was related to changes in clinical practice remains to be established. Since the onset of the COVID-19 pandemic, the US and countries worldwide have seen variation in reported incidence, testing patterns, case fatality rates, investigations of novel treatments, and clinical practice approaches. A few, [1] [2] [3] but not all, 4 recent non-peer reviewed reports suggested that the case fatality rate of COVID-19 infection is decreasing, and care is changing. However, there is an absence of published reports on large series of hospitalized COVID-19 patients, particularly with respect to temporal changes in clinical outcomes within the same data system. One key intervention is mechanical ventilation, initially thought to be best started early with more severe COVID-19 respiratory finding, notably hypoxemia. Over time, informal reports note less mechanical ventilation use; potential explanations for lower rates of COVID-19 mechanical ventilation and less mortality include: (i) changing demographics of patients; 5 (ii) more judicious use of starting and using mechanical ventilation; 6 (iii) more frequent use of anticoagulants; 7,8 and steroids; 9-12 (iv) changing prominent disease manifestations of patients; 13 (v) seasonal effects of temperature and humidity variation; 14 and (vi) potential changes in viral infectivity or pathogenicity. 15, 16 There have been temporal changes in COVID-19 testing in addition to no national testing strategy. Early in the pandemic, only those with symptoms and a higher pretest probability of COVID-19 exposure or disease were the focus of testing. Soon thereafter, COVID-19 testing increased in many, notably populations with higher frequency of potential infection (e.g., nursing home patients as well as health care professionals), and more recently increased screening of asymptomatic patients such as those with return to college campus activities, along with more . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted September 29, 2020. . https://doi.org/10.1101/2020.09.29.20203802 doi: medRxiv preprint intensive contacting tracing (and testing) of individuals likely to have been exposed to those carrying SARS CoV2 virus. [17] [18] [19] [20] We sought to examine clinical outcomes of hospitalized COVID-19 patients in a large health care system accompanied by an assessment of demographic and clinical profiles, along with changes in hospital-based clinical practice, that have occurred since the onset of the COVID-19 pandemic in the US. We used data routinely captured in the University of Pittsburgh Medical Center (UPMC) electronic medical record (EMR) systems. In brief, UPMC is a large academic medical center and insurer, housed principally in Pennsylvania. 21 The UPMC data system has detailed sociodemographic and medical history data, diagnostic and clinical tests conducted, surgical and other treatment procedures performed, prescriptions ordered, and billing charges on all outpatient and in-hospital encounters, with diagnoses and procedures coded based on the International Classification of Diseases, Ninth and Tenth revisions (ICD-9 and ICD-10, respectively). We linked the primary data sources using common variables (deidentified) within the UPMC data ecosystem aggregated in its Clinical Data Warehouse (CDW) that include: (i) Medipac, the admit, discharge and transfer registration and hospital-based billing system; (ii) Cerner, the inpatient electronic medical record (EMR) for relevant clinical information for bedded patients at a UPMC inpatient hospital; (iii) Epic, the UPMC EMR for ambulatory office visits owned by UPMC (Community Medicine Inc. and (University of Pittsburgh Physicians); and (iv) Aria, the EMR utilized in most ambulatory Cancer Centers at UPMC for both radiation . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted September 29, 2020. Of these, the study population consisted of 1,076 patients who tested positive for COVID-19 and were hospitalized at one of 21 UPMC hospitals (see Supplemental Table S1 for listing of hospitals). Our study received formal ethics approval by the UPMC Ethics and Quality Improvement Review Committee (Project ID 2882), the ethics/oversight body for ensuring patient confidentiality and consent (including waiver of consent) for analysis and dissemination of deidentified data within the UPMC system. Primary and Secondary Outcomes. The primary outcome for this analysis was a composite of in-hospital mechanical ventilation or mortality. We assessed in-hospital mechanical by the presence of a charge of mechanical ventilation within the Medipac billing software, specifically codes 94002 (first day of mechanical ventilation) and 94003 (each subsequent day of mechanical ventilation). Because some facilities used intensive care units for the care of COVID-19 patients regardless of illness acuity, we did not use intensive care unit admission as an outcome. We assessed in-hospital mortality by the discharge disposition of "Ceased to Breathe" sourced from the inpatient Medical Record System. Secondary outcomes included in-hospital mechanical ventilation or mortality individually and length of hospital stay (calculated by taking the difference between a discharge date time and the admission date time) available through billing within the Medipac system. Study . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted September 29, 2020. . https://doi.org/10.1101/2020.09.29.20203802 doi: medRxiv preprint investigators remained unaware of ascertainment of mechanical ventilation and mortality within the UPMC system during data collection and analysis. Explanatory Variables. For assessment of temporal changes and prior to analyses, we categorized the study analysis period into discrete epochs based on observed empirical changes in testing patterns in the UPMC system. We chose a 4-time period classification scheme (see . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted September 29, 2020. . https://doi.org/10.1101/2020.09.29.20203802 doi: medRxiv preprint We used propensity score methodology to compare in-hospital outcomes between epoch 4 versus epoch 1-3, adjusting for differences in presenting characteristics,. 22, 23 Logistic regression models were fit using hospital admission during epoch 4 as the dependent variable with stepwise selection (at p < 0.2) of measured pre-treatment explanatory variables. Resulting propensity scores (i.e. predicted probability of being in epoch 4 versus epochs 1-3) were the output and used as a continuous variable to control for confounding in Cox proportional hazards regression and logistic regression models of in-hospital outcomes. In sensitivity analyses, models were replicated with the use of inverse probability weighting (IPW), as well as 1:1 propensity score matching (PSM) with a maximum propensity score probability difference of 0.01. We set the comparison alpha error at 0.05 without correction for multiple comparisons. In stratified analyses, we examined potential effect modification by patient age using 3 groups defined as: (i) less than 60 years of age; (ii) 60 to less than 75 years of age; and (iii) 75 years of age and older. We did not impute missing values in any of the analyses. Methods and results are reported in accordance with The REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) statement 24 (see Supplemental Table 2 ). The mean number of hospital admissions was 9.9 per day during epoch 4, which was 2to 3-fold higher than average daily admission during the earlier epochs (see Supplementary Figure S1 ). The overall rate of COVID-19 testing performed in epoch 4 also was higher ( Figure 1 ). Presenting characteristics of patients hospitalized with COVID-19 are listed in Table 1 . The mean patient age was similar across the 4 epochs (ranging from 63.0 to 65.2 years), as was . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Table 4 ). In propensity adjusted logistic regression models, RRs (expressed as odds ratios) of being discharged alive within 3, 5, or 7 days of admission were higher in epoch 4 patients compared to epoch 1-3 patients, ranging from an estimated 1.6-to 1.7-fold higher odds of shorter length of stay in epoch 4 patients (Table 4 ). In sensitivity analyses using 1:1 propensity matched group patients (epoch 4 vs. epochs 1-3), presenting patient characteristics were similar (see Supplemental Table S3) , and Cox and logistic regression estimates favoring better in-hospital outcomes in epoch 4 patients were similar (see Supplementary Table S4) . Outcome results using propensity inverse probability . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. Relative Risks of In-Hospital Outcomes by Age. As seen in Table 5 is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted September 29, 2020. . https://doi.org/10.1101/2020.09.29.20203802 doi: medRxiv preprint not appear to be attributable to differences in baseline characteristics of hospitalized COVID-19 patients, which were minimal and controlled for statistically with multiple analytic approaches. In our large, representative health care system, we observed marked temporal changes in clinical management of patients, The most notable changes in clinical management of patients in the most recent interval was s higher use of dexamethasone (and glucocorticoids overall), higher use of anticoagulants, earlier initiation of dexamethasone, glucocorticoids as a class, and use of anticoagulants. The observed treatment strategy of earlier and significantly more frequent use of dexamethasone in patients in the UPMC system likely was triggered by the July 2020 release of the RECOVERY trial, 9 which showed lower 28-day mortality in patients who were receiving oxygen with or without invasive mechanical ventilation, and who received dexamethasone as compared to placebo. Our observed practice pattern of more frequent use of steroids is also consistent with a recently published (September 2, 2020) meta-analysis of six trials involving random assignment of different steroids (dexamethasone, hydrocortisone, methylprednisolone) compared to placebo, and approximately 30% lower risk of 28-day mortality among patients treated with steroids. 25 Practice. An obvious question that arises from the present analysis is to what extent did the recent changes in clinical practice (e.g. greater use and earlier initiation of steroids and anticoagulants) lead to overall lower rates of mechanical ventilation and hospital mortality, as well shorter length of hospital stay? Unfortunately, this type of question is difficult to answer with an observational dataset. Specifically, in non-randomized settings, there is potential confounding by indication (indication bias) when comparing different treatment approaches. It is nearly impossible to determine whether a given treatment approach was initiated a priori, as opposed to preferentially in response to patient disease severity and/or . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted September 29, 2020. . https://doi.org/10.1101/2020.09.29.20203802 doi: medRxiv preprint hospital course, thereby rendering comparisons of such treated patients to those who did not receive the treatment as potentially biased. 26, 27 As recently articulated, we believe caution is wise and causality uncertain though possible with these observational data. 28 Our data provide a rationale for potential conduct of new RCTs, particularly pragmatic and adaptive types 29 of trials that can quickly study questions, such as the timing in which steroids and anticoagulants are administered among COVID-19 hospitalized patients. Multiple potential explanations exist for these apparent differential results. First may be the clinical decision for potential use of mechanical ventilation. A report of 5,700 COVID-19 patients admitted to 12 hospitals in New York showed high rates of mortality for mechanically ventilated patients over the age of 65. 30 These findings comport with indications that some physicians may be reluctant to initiate mechanical ventilation in elderly patients as compared to younger patients who typically present with fewer comorbidities, (e.g. 31 ) potential medical ethics arguments made for rationing use of ventilators by age, 32,33 and from the patient perspective, possibility that some elderly patients (coupled with family member input) may be particularly likely to expressly state orders against the use of mechanical ventilation. (e.g. 34-36 ) These types of competing influences may blur assessment of the true clinical risk of mechanical ventilation in elderly COVID-19 hospitalized patients. With respect to less indication in our dataset of a mortality reduction in recently treated elderly COVID-19 patients, these findings may represent the greater challenge and complexity in treating elderly hospitalized COVID-19 patients who may be frail and have extensive comorbidities. This is consistent with reports of age being an . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. The copyright holder for this this version posted September 29, 2020. . https://doi.org/10.1101/2020.09.29.20203802 doi: medRxiv preprint independent risk factor for COVID-19 mortality 37 as well as for higher levels of inflammatory dysfunction markers and weakened immune response 38 . heterogenous patient population (e.g. enhances generalizability), standardized collection of realworld data and algorithmic coding of variables harmonized in a clinical data warehouse collected for non-research purposes (e.g. reduces potential reporting and ascertainment bias), and access to a very large battery of sociodemographic, medical history, medication use, and clinical practice and outcome variables available for analysis. A limitation is that we extracted all variables from the EHR of a multisite health care system, making fidelity concerns persist. All current data are observational and cannot determine causality. We also did not collect data to inform changes in host biology or viral pathogenesis over time, and we did not attempt to assess other external factors, such as seasonal effects of temperature and humidity variation, and possible patientspecific directives against the use of mechanical ventilation. Lastly, the present analysis includes a small percentage of hospitalized patients (<13%) enrolled in clinical trials, including potential randomization to either hydroxychloroquine, steroids, immunomodulators, convalescent plasma, or placebo. While a potential impact, we think that effect is modest if at all present. Recently treated hospitalized COVID-19 patients in our large health care system have overall lower rates of mechanical ventilation/in-hospital mortality and shorter length of hospital stay compared to earlier intervals. The extent to which these improved patient outcomes are related to recent changing clinical practice (i.e. greater and earlier use of steroids and anticoagulants) is unknown but an opportunity for rigorous future controlled trials. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted September 29, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted September 29, 2020. . CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted September 29, 2020. . https://doi.org/10.1101/2020.09.29.20203802 doi: medRxiv preprint CC-BY-NC-ND 4.0 International license It is made available under a is the author/funder, who has granted medRxiv a license to display the preprint in perpetuity. (which was not certified by peer review) preprint The copyright holder for this this version posted September 29, 2020. . https://doi.org/10.1101/2020.09.29.20203802 doi: medRxiv preprint COVID-19 case mortality rates continue to decline in Florida Is coronavirus becoming less deadly Declining COVID-19 case fatality rates across all ages: analysis of German data Is Covid-19 growing less lethal? The infection fatality rate says 'no' First Opinion COVID-Net: A weekly summary of US COVID-19 hospitalization data Respiratory support in COVID-19 patients, with a focus on resource-limited settings Effects of temperature variation and humidity on the death of COVID-19 in Wuhan Second coronavirus strain may be more infectiousbut some scientists are skeptical Tracking changes in SARS-CoV-2 spike: evidence that D614G increases infectivity of the COVID-19 virus Testing strategy for coronavirus (COVID-19) in high-density critical infrastructure workplaces after a COVID-19 case is identified Universal and serial laboratory testing for SARS-CoV-2 at a long-term care skilled nursing facility for veterans July brings major changes for coronavirus testing and tracing of college athletes The central role of the propensity score in observational studies for causal effects An introduction to propensity score methods for reducing the effects of confounding in observational studies The REporting of studies Conducted using Observational Routinely-collected health Data (RECORD) statement Association between administration of systemic corticosteroids and mortality among critically ill patients with COVID-19. A meta-analysis Confounding by indication and related concepts