Demographic and consortium-level data

Over a span of 3 weeks, 96 total hospitals in the US (45), France (42), Italy (5), Germany (3), and Singapore (1) contributed data to the consortium. This was represented by 23 data collaboratives across these five countries. A total of 27,584 patients with COVID-19 diagnosis were included in the data set, with data covering January 1, 2020 through April 11, 2020. We collected 187,802 laboratory values and harmonized them across sites. Thirteen percent of sites submitted complete data sets that included values for each laboratory (39.1% for at least 13, and 43.5% for at least 12 of the 14 laboratory measurements). Breakdown of sites is shown in Table 1.

Table 1 Sites contributing data to the consortium.

Demographic breakdown by age and sex is shown in Fig. 1. Age distribution was different across countries and consistent with previously identified patterns. In particular, patients from Italy were more commonly over the age of 70 relative to other countries18. US institutions, despite representing a large number of active infections, had the lowest percentage of elderly patients diagnosed with COVID-19. Germany, with its three included hospitals and relatively small number of patients, was more similar to the US and had an increased number of male patients in the 50−59 age group.

Fig. 1

Percentages of patients along with 95% confidence intervals in each a country and b sex age groups.

We were able to capture the total number of identified new cases by site and date. To normalize across sites and countries with varying sizes, we reported 7-day average new case rate per 100K over time for each country normalized by the ratio between the inpatient discharge rate for each country and inpatient discharge rate for the 4CE sites in that country. As shown in Fig. 2, the adjusted 7-day average new case rates derived from 4CE consortium sites match reasonably well with those reported by JHU CSSE19 for Germany, US, and Singapore. The 4CE estimates were substantially higher for France and Italy, which could reflect the fact that 4CE sites in France and Italy were mainly concentrated in urban areas with high infection rates.

Fig. 2

Adjusted 7-day average new case rate per 100K by country based on 4CE contributors along with 95% confidence intervals compared to 7-day average new case rate collected by Johns Hopkins Center for Systems Science and Engineering (JHU CSSE).

Laboratory value trajectories

Our initial data extraction included 14 laboratory markers of cardiac, renal, hepatic, and immune dysfunction that have been strongly associated with poor outcomes in COVID-19 patients in previous publications. Laboratory trajectories of each hospital at the population level are presented online at Given limitations of data harmonization and space, we focused on five laboratory trajectories that represented inflammatory, immune, hepatic, coagulation, and renal function. As shown in Fig. 3, trajectory data were remarkably consistent for most institutions at day 1 (day when biological test was positive) with growing differences with continued hospitalization. Extensive data harmonization was performed, but we must emphasize that data from each day represented a potentially different population as patients were discharged, died, or laboratory studies were no longer performed. Data values from each hospital were an average of all studied patients a specified number of days after diagnosis.

Fig. 3: Laboratory tests representative of renal function (creatinine), systemic inflammation (C-reactive protein), coagulopathy (D-dimer), liver function (total bilirubin), and immune response (white blood cell count) visualized relative to date of diagnosis of COVID-19.

The top row shows weighted means and 95% confidence intervals across all patients. The second row shows unweighted country- (thick lines) and site-level (thin lines) means. The third and fourth rows show the number of patients and sites, respectively, contributing laboratory tests of each type on a given day.

Initial laboratory values were abnormal for all patients but were not indicative of organ failure. Major abnormal elevations were noted in CRP and D-dimer on the day of diagnosis. As the number of days from diagnosis progressed, remaining patients who were not discharged or died had, on average, worse values. For nearly all 14 tests, trends toward progressively abnormal values were consistent with worsening disease as inpatient stays continued. Most importantly, the initial values and trajectories were highly consistent with previous findings in studies from China20,21.

Creatinine, a measure of renal function and the most commonly performed laboratory test in our data set, was divergent over time across sites. Rising creatinine would be consistent with an increased proportion of ill patients with significant acute kidney injury over time. Hospitals in Italy, in contrast, did not see a dramatic rise in creatinine in their hospitalized population, while the small percentage of French and German patients remaining in the hospital for 2 weeks had clear signs of acute kidney injury. This may represent many underlying differences including a high mortality near the beginning of the hospitalization at Italian hospitals, severe right time censoring of remaining patients, or a difference in practice.

Total bilirubin, a measure of conjugation and function by the liver, was initially normal across most sites and showed increases—consistent with other hepatic laboratory tests—among persistently hospitalized patients. The other hepatic laboratory measurements, alanine aminotransferase (ALT) and aspartate aminotransferase (AST), were divergent across institutions and showed a more significant perturbation (see Hepatic impairment was not present in most patients on presentation and total bilirubin was only mildly elevated with continued hospitalization.

On average, white blood cell count (WBC), a measure of immune response, was within normal limits on presentation. Patients who remained in the hospital and survived had increasing WBCs over time without severe leukocytosis21. Lymphocyte and neutrophil count trajectories can be seen on the website. Procalcitonin and LDH were not commonly tested in the total patient population, but results are also online.

C-reactive protein, a measure of systemic inflammation, was notably elevated on presentation for all patients in the cohort with a very narrow confidence interval, consistent with previous findings20. Although it is of unclear importance, populations of patients who remained in the hospital, survived, and had ongoing laboratory testing showed improvements over time. Interestingly, despite a decreasing trajectory during the first week, a mild leukocytosis is observed in counterbalance during the second week. The implication may be that CRP is not predictive of ongoing hospitalization or CRP is being checked for patient populations where the laboratory is more commonly improving.

D-dimer, an acute phase reactant and measure of coagulopathy, was elevated across institutions and countries at presentation. It rose consistently in all populations who continued to be hospitalized with the disease. This was consistent with multiple studies that showed a prothrombotic element to the disease. Most importantly, changes were consistent across all sites and highly abnormal.

Data attrition

There was a large drop in the number of laboratory tests performed after the first day (see Fig. 4). Drop off in tests performed could be a result of death, length of stay, or change in frequency of data collection by the clinical team. From the maximum number of laboratory tests consistently checked on the first day after diagnosis, there was a rapid tapering in frequency of laboratory tests checked. These changes were particularly pronounced in Italy and France. We identified the number of days until the number of tests checked were 20% of their initial maximum value. Values for laboratory study for each day are presented on Results varied for each laboratory value and site. There was no obvious country-level pattern. Given that several of these tests, such as creatinine, were commonly checked nearly every day in ill patients, the implication was that patients were censored from the laboratory results because of discharge or death or changing practice pattern. Thus, for the purposes of this paper, we focused on trends in creatinine. We normalized the number of tests performed by day to the total performed on day 1. We then looked at the day when the number of tests performed was 20% of the maximum number performed for each site. For creatinine, for example, a drop-off in testing occurred between day 7 and 15 across institutions. Most patients who survived were likely discharged within this time frame or managed with much less monitoring. Further results can be found online.

Fig. 4

Drop-off in laboratory tests reported for creatinine relative to the day with the largest number of laboratory tests reported.

Differences at admission

There was greater between-hospital variation for laboratory test performance than between-country variation (see Fig. 5). At the time of diagnosis, there was significant variation between countries and between the hospitals in a specific country. There was no obvious signature presentation for a country for an individual laboratory value. For example, creatinine was a commonly performed laboratory study within a day of diagnosis. The overall standard deviation (SD) for creatinine values across countries was 1.47 while the SD within sites was 1.39. Standard deviation for countries was 1.64, 1.31, 1.13, and 1.62 within France, Germany, Italy, and the US, respectively. France was a special case as 39 hospitals were reported together by AP-HP and then compared with three hospitals in Bordeaux. This was an important finding that could suggest that laboratory values, as individual results, would not be able to fully explain the mortality differences between countries.

Fig. 5: Laboratory variation across countries and within sites for creatinine test performed within a day of diagnosis.

Values for mean value and standard deviation (SD) in creatinine level are shown. Large variations exist within sites and are often larger than the between country variation. The overall SD across countries was 1.47 while the SD within sites was 1.39. Standard deviation for countries was 1.64, 1.31, 1.13, and 1.62 within France, Germany, Italy, and the US, respectively.


Source link

Leave a Reply

Your email address will not be published. Required fields are marked *