Increased risk of death in covid-19 hospital admissions during the second wave as compared to the first epidemic wave. A prospective dynamic cohort study in South London, UK.

Objective: To assess whether mortality of patients admitted for covid-19 treatment was different in the second UK epidemic wave of covid-19 compared to the first wave accounting for improvements in the standard of care available and differences in the distribution of risk factors between the two waves. Design: Single-centre, analytical, dynamic cohort study. Participants: 2,701 adults ([≥]18 years) with SARS-CoV-2 infection confirmed by polymerase chain reaction (PCR) and/or clinico-radiological diagnosis of covid-19, who required hospital admission to covid-19 specific wards, between January 2020 and March 2021. There were 884 covid-19 admissions during the first wave (before 30 Jun 2020) and 1,817 during the second wave. Outcome measures: in-hospital covid-19 associated mortality, ascertained from clinical records and Medical Certificate Cause of Death. Results: The crude mortality rate was 25% lower during the second wave (2.23 and 1.66 deaths per 100 person-days in first and second wave respectively). However, after accounting for age, sex, dexamethasone, oxygen requirements, symptoms at admission and Charlson Comorbidity Index, mortality hazard ratio associated with covid-19 hospital admissions was 1.62 (95% confidence interval 1.26, 2.08) times higher in the second wave compared to the first. Conclusions: Analysis of covid-19 admissions recorded in St. Georges Hospital, shows a larger second epidemic wave, with a lower crude mortality in hospital admissions. Nevertheless, after accounting for other factors underlying risk of death for covid-19 admissions was higher in the second wave. These findings are temporally and ecologically correlated with an increased circulation of SARS-CoV-2 variant of concern 202012/1 (alpha).


35
Since its emergence in December 2019, the spread of SARS-CoV-2 has increased exponentially leading to the 36 declaration of a pandemic by the World Health Organization (WHO) on 11 March 2020, marking the 37 beginning of an outbreak that has posed immense challenges for health care systems across the globe [1]. 38 The first confirmed case in the United Kingdom (UK) was registered on 31 January 2020. At the beginning of 39 March 2020, the growing transmission rates lead to the introduction of a series of control measures that 40 escalated to a full national lockdown (23 March 2020). This was subsequently followed by a drop in 41 transmission and hospitalisation rates with restrictions being eased over the summer months (Jun -Aug 42 2020). However, in October 2020 infections began to increase again leading to a second wave of covid-19 43 cases. The implementation of a second lockdown (04 November 2020) followed by tiered control measures, 44 in place until the beginning of March 2021, were needed to reduce the transmission rates again [2]. As of 01 45 May 2021, the UK has recorded 4,418,819 confirmed cases, 463,485 hospital admissions, and 127,571 46 deaths [3]. 47 Variables 105 The outcome variable was in-hospital covid-19-associated mortality, ascertained from clinical records and 106 Medical Certificate Cause of Death (MCCD). The main explanatory variable for this analysis was covid-19 107 wave, and 31 June 2020 used as cut-off to separate both waves. First wave was used as baseline/reference. 108 Covariates of interest for this analysis included demographics ( Oxygen requirement was a dichotomous variable indicating whether the maximum FiO2 (Fraction of Inspired 117 Oxygen) was over 21%. ICU admission was defined as covid-19 pneumonitis admitted to ICU. Symptoms at 118 admission were respiratory or wider infective symptoms at time of presentation. The CFS level was collected 119 on a nine-point ordinal scale to assess frailty within two weeks of admission (1 being very fit, 2 well, 3 120 managing well, 4 vulnerable, 5 mildly frail, 6 moderately frail, 7 severely frail, 8 very severely frail, and 9 121 terminally ill); but to avoid sparsity categories 7 to 9 were grouped [18]. CFS was expanded to include all age 122 groups excepting those patients with disabilities which rendered it inappropriate. CCI is a widely used 123 comorbidity summary measure, based on age and a predefined number of conditions with an assigned 124 integer weight representing the severity of each condition; for this analysis scores of 8 or more were 125 grouped in one category [19]. All scores were calculated by clinicians experienced in the use of scales. Four 126 categories under social history reflected the level of assistance required for daily activities. 127

Statistical methods 128
The distribution of covariates was assessed for the entire cohort and across waves. Mortality rates and 129 person-time of observation were calculated for the main exposure groups and all covariates of interest. The 130 strength of the association was quantified using incidence rate ratios (IRR), and the statistical significance 131 using 95%CIs and p-values. Survival across the different waves was explored using time-to-event analysis and 132 log-rank to test the significance of the difference between the survival curves. 133 Cox regression was used to estimate the effect of wave on mortality adjusting for multiple covariates. The 134 proportional hazard assumption was explored graphically and by testing for a zero slope in Schoenfeld 135 residuals. Follow-up time was stratified using lexis expansion and to minimise bias, intervals were created so 136 they would contain the same number of events. The assumption of proportionality was supported. 137 A causal model was built using a stepwise backward approach where (non-forced) pre-defined covariates 138 were retained in the model unless there were problems with multicollinearity. Age and gender were 139 considered a priori confounders (forced variables). Age was fitted using restricted cubic splines, with knots 140 positioned so numbers of events between knots were approximately equally distributed. The full model 141 . CC-BY 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) The copyright holder for this preprint this version posted June 12, 2021. ; https://doi.org/10.1101/2021.06.09.21258537 doi: medRxiv preprint included age, gender and all variables found to confound the crude association between wave and mortality 142 (non-forced variables). A change in the magnitude  5.5% was considered an indication of confounding. ICU 143 admission was included in the model as a non-forced variable regardless of the degree of confounding of the 144 main association. Problems will multicollinearity on the main effect in the full model, were resolved using 145 RMSE (Root Mean Square Error) reduction for backward deletion of non-forced variables. The RMSE for the 146 full model was used as reference for each step; and RMSE for each reduced model was calculated as √[〖〖 147

148
Following the same methodology, we carried out a sub-analysis among those requiring ICU admission. Data 149 management and statistical analysis were carried out using R. 150

Governance and ethics 151
This study was approved by the Health Research Authority (20/SC/0220). This manuscript follows the 152 STROBE statement for reporting of cohort studies. 153 154 . CC-BY 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 distribution of characteristics at admission for the entire cohort and across waves is shown in Table 1. 161 Covid-19 patients admitted during the second wave, were more likely to be younger, with patients aged 40 162 to 60 years being more prevalent in the second wave (495, 27.2%) and patients aged over 80 years being 163 more prevalent in the first wave (273, 30.9%). The distribution of sex was similar for both waves with males 164 being overrepresented (57.6%). During the second wave, patients were more likely to lead an independent 165 life (1,101, 61.1%) or have some level of family assistance (369, 20.5%); intermediate levels of frailty (3 166 managing well, 4 vulnerable, 5 mildly frail) were also more prevalent in the second wave (1,103, 60.7%) than 167 in the first wave (392, 44.3%). The proportion of severely frail patients (CFS 7-9) admitted was lower in the 168 second wave (197, 10.8% vs. 164, 18.6% in the first wave). Admissions scoring 0-3 in CCI were more 169 prevalent during the second wave (891, 49.0% vs. 386, 43.7%); the reverse occurred for CCI scores 4-5 and 170 over 6 (29.3% and 27.0% respectively for first wave, vs. 25.3% and 25.7% for second wave). Absence of 171 respiratory or wider infective symptoms at onset (initial diagnosis through PCR) was more prevalent in the 172 second wave (361, 19.9%) compared to the first wave (50, 5.7%). 173 The distribution of medical interventions after admission are listed in Table 2. The prevalence of admitted 174 patients requiring oxygen during admission was similar in both waves: 76.3% (668) in the first wave and 175 73.5% (1,328) in the second wave. The use of HFNO/CPAP was more prevalent in the second wave (400, 176 22.2%) than during the first wave (81, 9.3%), whilst invasive ventilation was more prevalent in the first wave. 177 The distribution of patients requiring ICU admission had a similar distribution across waves (23.1%). The use 178 of dexamethasone, Remdesivir and Tocilizumab was almost exclusively during the second wave. During the 179 first wave there was some use of Tocilizumab (23, 2.6%) and dexamethasone (58, 6.6%), both used in the 180 context of clinical trials (with a few cases additional cases of compassionate use for Tocilizumab). 181 182 A total of 752 patients died over the total time at risk (40,777 person-days); 297 (33.6%) deaths occurred 183 during the first wave and 455 (25.3%) during the second wave. The median time of follow-up for those 184 discharged was 10 days (IQR: 5-22 days) and for those who died was 11 days (IQR: 5-19 days). We found no 185 differences in the overall distribution of the follow-up time across waves. Among those discharged, 186 admissions with lengths of stay (LOS) over 35 days were similar across waves (11.9% for the first wave and 187 11.1% for the second); LOS between 0-7 days were more prevalent in the second wave (562, 41.8%) than in 188 the first wave (188, 32.0%). The median probability of survival was 29 days (95%CI 30-41 days) for the first 189 wave, and 37 days (95%CI 32-47 days) for the second. 190 . CC-BY 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)
The copyright holder for this preprint this version posted June 12, 2021.  Table 3 shows the crude and adjusted IRR (IRRa) for the effect of wave on mortality on the same set of 206 observations. The strongest confounders of the association in this cohort were dexamethasone, oxygen 207 requirement, symptomatic at admission, CCI, and HFNO/CPAP. In all cases, removing the effect of the 208 confounder (IRRa) retained the protective effect of second wave to different degrees. Mortality was 43% 209 (95%CI 71%, 52%) lower in the second wave compared to the first wave when adjusting for the effect of 210 dexamethasone; and 16% (95%CI 3%, 27%) lower when adjusting for oxygen requirement. Thus, oxygen 211 requirement is acting as partial positive confounder whereas dexamethasone is acting as a negative 212 confounder in this cohort. 213 In the multivariable analysis, the hazard of death during the second wave was 1.62 times higher (95%CI We summarised the distribution of baseline characteristics at admission and medical interventions across 233 waves for the entire study cohort (Table 1). During the second wave, younger admissions with moderate 234 levels of frailty/CCI were more prevalent, compared to either older and/or frailer patients in the first wave. 235 In addition to this, during the second wave, we observed an increase of covid-19 specific treatments as trial 236 data emerged for the use of dexamethasone, Remdesivir and Tocilizumab. 237 The multivariable analysis attempted to account for all the available factors unequally distributed across 238 waves and also associated with mortality (while avoiding multicollinearity in the model). We found a 1.62-239 fold increase in the hazard of death (95%CI 1.26, 2.08), after controlling for the effect of age, sex, 240 dexamethasone, oxygen requirement (maximum FiO2>21%), symptoms at admission and CCI. dexamethasone. Both variables were included in the model as the level of uncontrolled confounding 254 reduced was larger than the error introduced due to collinear effects. After accounting for the effect of age, 255 sex, dexamethasone, oxygen requirement, symptoms at admission, CCI and wave, dexamethasone reduced 256 the hazard of death in this population of patients by 53% (95%CI 40%, 63%). 257 We further explored the effect of wave on mortality on the subpopulation of patients admitted to ICU, i.e., 258 the most severe covid-19 patients. All these patients had oxygen therapy so, this variable was not a factor in 259 the main model. Within this sub-group of patients, the hazard of death during the second wave was also 260 larger than in the first wave (HR: 2.00, 95%CI 1.10, 3.62) after accounting for the effect of age, sex, 261 . CC-BY 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)
The copyright holder for this preprint this version posted June 12, 2021. ; https://doi.org/10.1101/2021.06.09.21258537 doi: medRxiv preprint dexamethasone, Remdesivir, Tocilizumab and HFNO/CPAP. This further supports the observation that risk of 262 death in covid-19 hospitalised patients was higher in the second wave compared to the first wave, when 263 differences in the standard of care and the characteristics of the patients were taken into account. 264 There is some evidence that VOC 202012/01 (alpha) is associated with increased risk of death [12-14]; but 265 since S-gene target failure (SGTF) detection or genomic sequencing data were not available for this study 266 population, attributing our observation of increased in-hospital mortality to variant VOC 202012/01 would 267 largely depend on the acceptability of the assumption that said variant was dominant in our catchment area. 268 This might not be an unreasonable assumption as prevalence of SGTF (associated with this new variant), was 269 already at 5.8% at the beginning of November 2020, increasing sharply to reach 94.3% at the end of January 270 2021 [13]. 271

Strengths and limitations 272
This was a large analytical cohort study comparing groups of patients at different points in time. The overall 273 goal was to investigate if different standards of care and possible changes in the natural history of the 274 disease (attributed to changes in SARS-CoV-2 variants), had an impact on in-hospital mortality. We included 275 all patients admitted to covid-19 wards for treatment. 276 All variables used in this study were extracted prospectively from electronic medical records ensuring data 277 collected were the same across waves. The majority of the data were collected by experienced respiratory 278 and ICU clinicians, and although some data inconsistencies were rectified early during data management, 279 misclassification of covariates due transcription errors cannot be ruled out. Laboratory variables such as 280 oxygenation parameters were obtained through the informatic department, but due to the limited quality of 281 the electronic records, data were inconsistent and, in many cases, missing. We dichotomised this variable 282 (FiO2) in an effort to reduce measurement error, but the coarse categorisation of oxygenation parameters 283 into a dichotomous variable is likely to have introduced residual confounding. 284 Outcome and date of outcome were collected separately and ascertained from MCCD (available for 752 285 deaths, 94.1%). The number of deaths we observed during the first wave is consistent with numbers 286 previously reported for the same catchment area and period [23]. However, it has been observed that 287 during the first epidemic wave in the UK there was a larger mortality within care homes [2], so it is possible 288 that we have underestimated the number of deaths in the first wave. This differential misclassification of 289 outcome could have led to an overestimation of the effect of the second wave. In addition, temporal effects 290 could also have explained some of the observed differences between waves, as fatality rates are known to 291 be higher during winter months, when the second wave unfolded. Overall, there was a good level of data 292 completeness with only BMI observing large numbers of missing values. 293

Generalisability 294
This study was looking at an overall population of hospitalised adults with covid-19 in a large reference 295 teaching London hospital. Findings are only generalisable to inpatient population. 296 . CC-BY 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 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)
The copyright holder for this preprint this version posted June 12, 2021. ; https://doi.org/10.1101/2021.06.09.21258537 doi: medRxiv preprint  . CC-BY 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) The copyright holder for this preprint this version posted June 12, 2021. ; https://doi.org/10.1101/2021.06.09.21258537 doi: medRxiv preprint