COVID19 Severity Prediction and Health Services Research Evaluation

Background: One of the fundamental problems of this epidemic is determined by the high
percentage of SARS-CoV-2 infected patients who present rapid clinical deterioration that
makes them need care in critical units. Identifying which factors are related to these more
severe conditions would allow us to assess whether preventive or therapeutic measures can be
put in place in advance or to better plan the services to be provided to these patients,
either in this wave of the pandemic or in those that may occur in the future.

Objectives: This project aims to create stratification scales of the risk of poor evolution
in patients infected by SARS-CoV-2, defined as the appearance of clinical deterioration,
ARDS, sepsis, SRIS, septic shock or death. Additional goals are: 1.-Evaluate the
accessibility and equity that these patients have had in the different care processes,
diagnostic and therapeutic procedures, with special interest in patients who came from
residences, by age, gender or geographic origin. 2.-Evaluate the effectiveness of different
therapeutic schemes that have been used in this pandemic. 3.-Evaluate the effectiveness of
different diagnostic tests used to predict the poor evolution of these patients 4.- Evaluate
the real costs associated with the treatment of hospitalized patients with COVID-19 Methods:
The information will be extracted from the electronic medical record mostly, but will have to
be done manually for certain fundamental parameters of prediction (clinical manifestations,
date of onset of symptoms and duration of symptoms, and epidemiological history). Statistical
analysis: Logistic regression/survival models/artificial intelligence algorithms will be
created for the prediction of poor evolution of patients with CoVid-19.

Two samples are included: 1.-All people COVID19 positive from the Basque Country (around
18768 people); 2.-Patients admitted for COVID 19 in the centers participating in the study
during the first wave of the pandemic, until May 31, will be included (in the case of the
Basque Country, some of these patients will come from the population sample #1 described
before). If there were new waves of a certain entity (more than 100 admissions in a month per
center), this information would also be collected later. With the information the
investigators have so far, the investigators see that the investigators would have between
6000-7000 to select. Later, patients from the autumn wave would be collected, if it were
given, until the end of May 2021, due to greater temporal similarity with the first wave.

Sampling: The information to be reviewed from the medical record will be collected from the
first wave of the pandemic between March-May 2020, where a random sampling will be carried
out . For the second wave of autumn-winter of 2020-2021, a random sample of patients will
also be collected, enough to meet the estimated sample size for this second wave. If not, the
sample size will be completed with patients from the first wave.

VARIABLES: Exposure: 1.-Sociodemographic data: Age, gender, residence (yes / no), country of
origin. 2.- Personal history: associated diseases; Basal treatments, etc. 3.-History of the
disease 4.-Physical examination at home or AP. 5.-Hospital history: symptoms on arrival at
the emergency department, vital signs, signs and physical examination, Laboratory tests,
chest radiography pattern, CAT pattern, established treatments, ICU data.

Result: Clinical impairment: Dyspnoea at rest, Development of ARDS, sepsis, SIRS, shock, ICU
admission, Death (date). Relief of symptoms, days until the absence of disease, death.

Follow-up (6 months): Readmissions, New diagnoses, Complications, Biomarkers of fibrogenesis,
Results of the diagnostic procedure (radiographs, MRI, CT), Death (with date and cause) Costs
(index and 6 months income): Emergency or programmed admission; number of days of admission
(in each of the Units / Plants / ICU / Emergencies); laboratory tests (number and type);
number of days in which respiratory support was required; treatments used throughout the stay
(drug, dose, dosage, duration); diagnostic procedures (radiographs, MRI, CT, etc.) performed
during the study period; surgical procedures performed; external consultations (number and
Service); day hospital (number and procedures); AP and home visits (related to COVID-19) DATA
COLLECTION METHODS: Manual data extraction will be carried out by reviewers under the
supervision of each PI per center. All the collected data will be entered in the RedCap
database. Once the information is extracted, a common database will be created for subsequent
analysis.

STATISTIC ANALYSIS. The study unit will be the patient. A descriptive analysis of the entire
sample will be carried out. A univariate analysis will be performed to determine potential
factors or variables related to the outcome variables of interest. In the multivariate
analysis, different models will be carried out according to the dependent variable of
interest. In the case of dichotomous dependent variables, logistic regression models will be
used. Statistical significance will be assumed when p <0.05 and all analyzes will be
performed using SAS v9.4 and R statistical software. Also, the prediction of the variables
will be evaluated individually by measuring the statistical correlation between each variable
and the poor evolution; and collectively looking at the ability to predict the bad evolution
from combinations, which will be obtained by generating Association Rules between variables
from the underlying statistical relationships.

The analysis of the comparative effectiveness between the different treatment options that
have been observed will be carried out by intention to treat. In addition to descriptive
statistical techniques, a time-to-event (mortality) survival analysis will be performed using
multivariate Cox proportional hazards regression, and a parametric survival analysis with the
corresponding distribution (Weibull, etc.) together with an estimate. of average survival.
For the evaluation of comparative effectiveness, propensity score techniques will be used to
create comparable treatment groups by adjusting baseline covariates by inverse weighting of
treatment probability. Additionally, and because it is foreseeable that there will be
multiple treatment groups, the specific estimation procedure called generalized boosted
models will be applied.

For the analysis of cost data, both for the analysis of associated variables and for a cost
comparison objective, GLM regression techniques will be used with the type of distribution
that best fits the data (using the Modified Park Test ), although preferably the gamma and
logarithm family will be used as the link. The data will be analyzed with the Stata v14.2
program.

ETHICAL AND CONFIDENTIALITY ASPECTS. The project has been evaluated by the research
commissions and the Research Ethics Committee with Medicines (CEIm), where it has been
approved. The laws on personal data will be followed (RGPD 2018) All information will be
treated in an absolutely confidential manner.

Expected results: A prognostic stratification tool based on predictive models of poor
evolution in CoVid-19 infection: clinical deterioration and development of ARDS, SRS, sepsis,
and/or septic shock and/or death. This tool will help guide the most appropriate clinical
management of patients, mainly those with the most severe presentations that may require
attention in critical care units. Additionally, purposes of this study are alsoto provide
information on the variability and costs in the provision of health care that may have been
given, both in the use of diagnostic tests and in the use of different therapeutic options
and also in the results finally obtained. The investigators seek to identify problems in the
accessibility of different groups (elderly, people in residences, by gender, higher level of
comorbidities, immigrants ...), and that can help us identify problems in equity in access to
health services.