Hospital referrals, exclusions from hospital care, and deaths among long-term care residents in the Community of Madrid during the March-April 2020 COVID-19 epidemic period: A multivariate time series analysis

doi:10.21203/rs.3.rs-3793981/v1

Download PDF

Research Article

Hospital referrals, exclusions from hospital care, and deaths among long-term care residents in the Community of Madrid during the March-April 2020 COVID-19 epidemic period: A multivariate time series analysis

https://doi.org/10.21203/rs.3.rs-3793981/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 14 Aug, 2024

Read the published version in BMC Geriatrics →

You are reading this latest preprint version

Background:

From March 6 to April 7, 2020 the Community of Madrid (CoM), Spain, issued interventions in response to the COVID-19 epidemic, including hospital referral triage protocols for long-term care facility (LTCF) residents (March 18–25). Those with moderate to severe physical disability and/or cognitive impairment were excluded from hospital referral. This research assesses changes in the association between daily hospital referrals and the deaths of LTCF residents attributable to the triage protocols.

Methods:

Daily hospital referrals and all-cause mortality from January to June 2020 among LTCF residents were obtained. Statistically significant changes in LTCF resident hospital referrals, and daily deaths in-hospital and in-LTCF were identified when testing for breaks and regimes in the time series. Multivariate time series analyses were conducted to test changes in the associations between LTCF resident hospital referrals with in-hospital and in-LTCF daily deaths as well as with daily deaths in the CoM population aged 65 + at the time triage protocols were implemented.

Results:

Among LTCF residents, hospital referrals declined sharply from March 6 to March 23, 2020. Increases in daily deaths at LTCFs occurred from March 7 to April 1 followed by a decrease reaching pre-epidemic levels after April 28. The daily ratio of in-hospital deaths to in-LTCF deaths reached its lowest values from March 8 to April 13, 2020. The four versions of the triage protocol, published from March 18 to March 25 had no impact on further changes in the association of hospital referrals with daily deaths of LTCF residents in-hospital or in-LTCF. Daily deaths in the CoM population aged 65 + were the main driver of daily deaths of LTCF residents.

Conclusions:

During March-April 2020, while hospital referrals decreased because of CoM governmental interventions, daily deaths of LTCF residents within LTCFs increased. Increases in hospital ICU bed occupancy and non-ICU bed occupancy by non-residents of LTCF suggest that the CoM government chose to protect hospitals from collapse by restricting hospital referrals from LTCFs without concurrently implementing crisis standards of care within LTCFs. CoM triage protocols sanctioned the already existing restrictions on hospital referrals of LTCF residents set by the CoM government.

COVID-19

Mortality

Long-term care facility

hospital referrals

Triage protocols

Multivariate time series

Decreases in hospital-based emergency care and planned hospital admissions were observed across numerous high-income countries during the COVID-19 pandemic [1]. Hospitalizations in ICU (Intensive Care Unit) and non-ICU (Non-Intensive Care Unit) beds more than compensated the decrease in non-COVID-19 hospital bed occupancy by COVID-19 inpatients [1]. Low to moderate levels of hospital bed saturation were experienced in some of these countries [2]. The pressures originating from COVID-19 on the healthcare system in Spain were at least seven times higher on the Verelst et al. [3] pressure intensity scale than those in France, Switzerland, the UK, Belgium, Denmark, Luxembourg, and Sweden. Spain shared the same challenges related to the COVID-19 epidemic as other high-income countries. Nevertheless, Spain experienced the highest level of COVID-19 mortality among persons aged 65 + among twelve OECD countries [4]. Spain also had the highest long-term care facility (LTCF) resident mortality [4].

The Comunidad Autonoma de Madrid (Autonomous Community of Madrid – CoM) had the highest excess mortality of all comunidades autonomas (autonomous communities - ACs) in Spain in the March-April 2020 first wave of the COVID-19 epidemic [5, 6]. During March and April 2020, 19% of the older adults living in LTCFs in the CoM died [7, 8]. COVID-19 spread in LTCFs as in other countries [9–11]. In a study of data from the USA and 12 European countries, Aaolto et al. found a positive correlation between COVID-19 cases and deaths in LTCFs and the total population [5]. In a study of data from Catalonia [12], the cumulative incidence of COVID-19 mortality was associated with mortality in LTCFs. Mortality in the CoM LTCFs COVID-19 patients was the highest amongst all published studies in Spain with more than double the number of deaths occurring in LTCFs, compared with studies conducted in regions other than the CoM [13]. Among the policy decisions that may have contributed to these statistics, Koleva et al. [9] identified the protocols restricting hospital referrals of LTCF residents based on triage by public hospital geriatricians issued between March 18 and March 25, 2020 [14–17]

These triage protocols restricted hospital referrals of LTCF residents with moderate to high levels of disabilities and/or cognitive impairments [13, 14]. Their main goal, allegedly based on evidence, was to ensure coordination between LTCFs and hospitals regarding the provision of healthcare for LTCF residents with COVID-19. One of the secondary objectives was to prevent the collapse of the CoM healthcare system, specifically of the community’s hospitals. In effect, non-elective hospitalizations increased by 70% from March 14 to 24, 2020 – from approximately 1000 hospitalizations per day to approximately 1700. Most were COVID-19-related cases[18] (pages 48–50). In this context, decreasing hospital referrals of LTCF residents was aimed at reducing pressure on ICU and non-ICU bed occupancies.

Healthcare jurisdictions restricted hospital referrals during the first wave of COVID-19 with similar protocols. For example, the Government of Québec (Canada) also implemented triage protocols restricting access to hospital admission by LTCF residents. Hospital referrals were allowed for LTCF residents with severe COVID-19 symptoms and those in need of ambulatory and in-patient care not available in LTCFs. In a cohort of LTCF residents in Montréal, only 5.3% of LTCF residents with severe COVID-19 symptoms were hospitalized in the first wave of the epidemic - March 24, 2020 to July 9, 2020 [19]. Conversely, in the CoM’s triage protocols, the severity of disability and/or cognitive conditions, not the patients’ COVID-19 condition, restricted access to hospital referrals.

In Spain, Autonomous Communities (AC) have the responsibility and decision-making power for policies, funding, organization, and standards concerning LTCFs [20, 21]. LTCFs in the CoM are homes for residents that deliver personal, gerontological, and rehabilitation, but not medical, care. They are responsible for referring residents to healthcare providers as mandated by the Spanish National Health System. The AC responsibilities were confirmed on March 14, 2020, by the Spanish Government’s Royal Decree 463/2020 establishing a state of emergency in response to the epidemic [22]. AC were ultimately made responsible for extending healthcare provisions to LTCFs [22], including medical, nursing, and paramedical care. In addition, by this Royal Decree, referrals of LTCF residents in need of care not available in LTCFs to healthcare facilities able to provide appropriate levels of care were required (Article 5.2).

These triage protocols were the most politically challenged public health intervention in response to the epidemic in the CoM. Restrictions on hospital referrals of LTCF residents raised concerns regarding access to vital hospital care and the increased number of daily in-LTCF deaths. Thus, this research focuses on the CoM’s disability-based triage protocols implemented between March 18–25, 2020 restricting hospital referrals of LTCF residents [14, 15]. The objective of the study is to examine the association between the triage protocols, the enactment dates of the triage protocols and changes in daily hospital referrals and deaths of CoM LTCF residents in-hospital and in-LTCF.

The objective is examined using the three following approaches:

Government of the Community of Madrid (GCM) COVID-19 epidemic interventions in March-April 2020 and LTCF residents' daily hospital referrals and in-hospital and in-LTCF deaths were examined in conjunction.

Breaks in LTCF residents’ daily hospital referrals, and daily in-hospital and in-LTCF deaths due to the enactment of the triage protocols were examined using univariate time series analyses.

The extent to which LTCF residents' daily hospital referrals and in-hospital and in-LTCF deaths were associated with the March 18–25, 2020 triage protocols were probed using a multivariable data-generating model

Material

Data on daily hospital referrals and daily deaths by place of death (hospital or LTCF) of LTCF residents, as well as on daily deaths in the population aged 65 + in the CoM, were obtained. Data extends from January 5 to June 27, 2020, thus covering the COVID-19 epidemic's first wave and the pre- and post-COVID-19 periods. We define March-April 2020 as the COVID-19 epidemic period.

The Transparency Portal of the Community of Madrid (Portal de Transparencia de la Comunidad de Madrid), a government office, and the National Statistics Institute (Spain) provide public access to information upon request. Table 1 illustrates the data sources, populations, and periods covered in the analyses. Since daily deaths for all causes of LTCF residents occurring after hospital referrals were not available from the National Statistics Institute, we used data sets available from the GCM’s Ministry of Social Policies, Families, Equality and Fertility to estimate the number of LTCF resident in-hospital deaths for all causes in the COVID-19 and post-COVID-19 periods.

Insert Table 1 here

Table 1. Sources of hospital referrals and mortality data

Data for all-cause daily deaths of LTCF residents in hospitals were not available for the pre-COVID-19 period. Values were estimated using a linear regression where predictors were the post-COVID-19 daily hospital referrals, daily deaths of persons aged 65 + living in the CoM, and daily deaths in LTCFs. The predicted variable was the daily in-hospital deaths of LTCF residents in the pre-COVID-19 period. Durbin’s alternative test for autocorrelation rejected the hypothesis of serial correlation among residuals from the regression at the 0.05 level. Predicted values of in-hospital deaths of LTCF residents in the pre-COVID-19 period obtained from the regression were used in this study (available on request).

Six variables were constructed from the data files: 1) Daily deaths in the population aged 65 years + living in the CoM; 2) Total daily deaths of LTCF residents; 3) Daily deaths occurring at the LTCFs; 4) Daily in-hospital deaths occurring after referrals from LTCFs; 5) Daily referrals from LTCFs to hospitals; and 6) Hospital-to-LTCF mortality daily ratios.

Statistical procedures

Data visualization

Graphics were used to represent the observed time series for daily deaths, hospital referrals, and the sequence of administrative interventions, including the March 18–25 triage protocols. For descriptive purposes, the March-April 2020 period of the COVID-19 epidemic start date was set as February 25, 2020, the date of the first hospitalization of a suspected COVID-19 case in the CoM (Additional File 1). The end date was set on April 28, the date the Government of Spain published the Plan for the Transition to a New Normality [23].

Univariate analyses of changes in hospital referrals and daily mortality

Breaks result from shocks changing the direction of time series [24]. Each break is followed by a regime lasting until the next break. As an illustration, breaks and regimes for the hospital referrals of LTCF residents’ time series are plotted in Fig. 1. Changes in hospital referrals introduced by shocks are illustrated by two breaks (March 6, 2020 and April 1, 2020) and two regimes (March 6, 2020 to March 31, 2020 and April 1, 2020 to April 29, 2020). These correspond to the ascending and descending phase of the March-April period of the COVID-19 epidemic (all events in this paper occurred in 2020). These statistical analyses aim to identify the exact starting and ending dates of statistically significant (P ≤ 0.05) changes in the time series (the breaks) and the course of the time series between break dates (the regimes).

Insert Fig. 1 here

Figure 1. Examples of breaks and regimes:

The case of hospital referrals

The univariate analyses examine whether shocks on daily hospital referrals and daily deaths are statistically significant and whether they coincide with the date the triage protocols were issued.

The operational objective pursued by the triage protocols was to introduce changes – breaks and regimes – in LTCF residents’ hospital referrals. Due to implementation delays, the breaks may have occurred in the days following March 18, 2020, when the first protocol was issued. Hence, the exact break dates and length of the regime for hospital referrals are unknown. The exact starting and ending dates of statistically significant breaks and regimes in daily in-hospital and in-LTCF deaths during the COVID-19 epidemic are also unknown. The Ditzen et al. [24] procedures for unknown break dates have been used to identify breaks and regimes (see Additional File 2).

Increases in daily variations within a regime may be associated with turbulence or volatility in a time series [25]. Thus, a high volume of all-cause deaths during the COVID-19 epidemic may have occurred in conjunction with increases in the range of variation of unforeseen patterns of daily deaths during a regime leading to a high level of uncertainty in the flux and management of care to patients in hospitals and LTCFs.

This study used four tests to examine time series attributes [26] and to conclude on stationarity and volatility (see Additional File 2).

Multivariate analyses of hospital referrals and in-LTCF and in-hospital deaths

In this study, the foundational data-generating process was built considering critical sequences of events during the March-April 2020 period of the COVID-19 epidemic. Two sequences of events were considered as “impulses” which generated “responses” in other sequences of events. The first impulse was the increase in daily deaths in the CoM population aged 65+. The changes in daily hospital referrals and in-hospital and in-LTCF daily deaths were responses to this impulse (Fig. 2). Hospital referrals and in-LTCF and in-hospital deaths of LTCF residents were postulated as increasing in the ascending phase of the COVID-19 epidemic and decreasing in the descending phase. Thus, positive signs are shown alongside the arrows between hospital referrals and in-LTCF deaths (Fig. 2).

The second impulse consisted of the four triage protocols enacted by the CoM from March 18, 2020 to March 25, 2020. The first protocol was issued approximately four weeks after the hospitalization of the first suspected COVID-19 case in the CoM on February 25, 2020. In line with one of the triage protocols' secondary goals, hospital referrals decreased after their introduction – see the negative sign in Fig. 2. As the number of LTCF residents referred to hospitals decreased, they were locked down in LTCFs, and the number of LTCF residents dying on premises increased (see the positive sign in Fig. 2). Concomitantly, the number of LTCF residents dying in hospitals decreased (see the negative sign in Fig. 2).

Insert Fig. 2 here

Figure 2. The data-generating model

From Fig. 2, a multivariate model with three equations was designed to obtain estimates of daily hospital referrals and daily deaths of LTCF residents in-LTCF and in-hospital (Fig. 2):

Equation 1: Decreases in LTCF residents’ hospital referrals were responses to the two foundational impulses.

Equation 2: Increases in daily in-LTCF deaths were responses driven by the two foundational impulses and decreases in hospital referrals.

Equation 3: Decreases in daily in-hospital deaths among LTCF residents were responses driven by the two foundational impulses and by decreased hospital referrals of LCTF residents and increased daily in-LTCF deaths.

In the multivariate model, breaks and regimes were coded into three variables: shifts, pulses and ramps [24]. 1) Shifts, i.e. sudden and very short shocks in the time series, were coded "1" on the day of the event and 0 otherwise. 2) Pulses, i.e. non-varying shocks over time, were coded "1" on days included in the regime and 0 otherwise. 3) Ramps, i.e. temporary changes, were coded from 1 to T, increasing by one unit for each day of the associated regime until the end “T” of the regime [32].

Stationary and volatility tests were conducted to guide the choice of the multivariate time-series model [27]. Three models were considered as candidates: the vector autoregressive model (VAR), the vector error-correction model (VECM), and the MGARCH model. [33]. For these analyses, the MGARCH model was selected (See Additional File 2). Statistical analyses were run on Stata 15 [28].

As all data obtained from administrative files were daily events — deaths and hospital referrals — occurring to the population aged 65 + and to LTCF residents in the CoM without identifiable personal data, this study was not submitted to a research ethics committee.

(1) A narrative of GCM interventions in the March-April 2020 COVID-19 epidemic and of LTCF residents' daily hospital referrals and daily deaths in hospitals and LTCFs

As early as January 30 the CoM’s Directorate General of Public Health (DGSPCM-Dirección General de Salud Pública de la Comunidad de Madrid) issued a procedure for hospitalization of suspected cases of COVID-19. This proposal was rejected by the GCM (Additional File 1). The first hospitalization of a suspected case in the CoM occurred on February 25, 2020 [29]. Centralized management of hospital beds was initiated on March 7, 2020 (Additional File 1). Also, the three-phase Hospital Resource Elasticity Plan was promoted (see[18] pages 45–50). Necessary hospital resources were anticipated to increase alongside the increases in the number of COVID-19 inpatients and outpatients.

Restrictions on hospital referrals were issued in the first and third weeks of March 2020 (Fig. 1; Additional File 1). From March 8, the date of the lockdown of LTCFs, the GCM gradually introduced more severe restrictions on access to healthcare located outside of LTCFs (Additional File 1). In addition, there was evidence that LTCF managers encountered increasing barriers to access of their residents to hospitals [29–31]As daily in-hospital deaths of LTCF residents decreased while daily LTCF deaths increased, the ratios of the former to the latter started to decrease abruptly after March 8, 2020 (Fig. 3a), five days after the start of the GCM intervention period.

Insert Fig. 3 here

Figure 3. Times series for all-causes daily deaths, hospital referrals,

and GCM interventions

a. Daily deaths in the CoM for the population aged 65 + and those in LTCFs

b. LTCF residents' daily referrals and in-hospital and in-LTCF daily deaths

c. Hospital-to-LTCF ratios of daily deaths of LTCF residents

Compared to the pre-COVID-19 period, hospital referrals from LTCFs decreased abruptly from March 3, 2020 to March 17, 2020 (Fig. 3b; Table 2b). According to the GCM Ministry of Health action plan (March 12, 2020), LTCF residents were to be treated on-premises, and the medicalization of LTCFs was announced (Additional File 1). While the medicalization of LTCFs was never defined by the GCM Ministry of Health, in the 2020 GCM Ministry of Health Annual Report, guidelines for the medicalization of hotels included the ability to provide the means to control the spread of COVID-19 among residents and staff, to reduce necessary hospitalizations, and to increase the ability to meet necessary care. Organizational and clinical links with hospitals were also planned[18] (pages 79–80). Medicalization was defined in the May 6, 2020 order of the High Court of Justice of Madrid as providing LTCFs with medical and nursing staff, and the appropriate ways and means necessary to care for LTCF residents during an epidemic situation[30] However, the GCM Ministry of Health never implemented the medicalization of LTCFs [32]

Between March 18, 2020 and March 25 average daily in-LTCF deaths increased, reaching a ratio of 5.41 (Table 2a, b) compared to the January 5, 2020 - March 3, 2020 period. Hospital referrals reached their lowest levels (Fig. 3b; Table 2b). Meanwhile, the GCM’s “shock plan” (plan de choque) [33] was made public on March 26, 2020 (Additional File 1). The highest levels of daily deaths in the CoM population aged 65 + and in LTCF residents were observed between March 26, 2020 and April 7, 2020 (Fig. 3c; Table 2a). The daily in-LTCF death average peaked at 15 times the January 5, 2020 to March 3, 2020 period daily death average (Table 2b), while the daily death average of LTCF residents after hospital referrals was only 1.57 times higher (Table 2b). The gradual retreat of GCM administrative directives began on April 7, 2020 (Fig. 3b,c) when the ratios of daily deaths in the CoM population aged 65 + and in LTCF residents decreased (Table 2b). After the implementation of New Normalization Phase 1 on May 23, 2020 (Additional File 1), these ratios deepened below unity, except for deaths in hospitals (Table 2b). Finally, the daily hospital referral average in the January 5, 2020 to March 3, 2020 period was higher than that in the March 3, 2020 to April 6, 2020 period and even higher than in the May 23, 2020 to June 27, 2020 period (Fig. 3b; Table 2a).

Insert Table 2 here

Table 2. Descriptive statistics on hospital referrals and all-cause deaths in the CoM: population aged 65 + and LTCF residents

a. All-cause deaths and hospital referrals

b. Ratios of all-cause deaths in persons in LTCFs and persons aged 65 + in the CoM

(2) Univariate time series analyses of changes, vis-à-vis the enactment of the triage protocols, LTCF residents’ daily hospital referrals, and daily in-hospital and in-LTCF deaths, were examined.

Changes were modeled as breaks in the respective time series. Full results on breaks are available in Additional File 2, Supplementary Table 2.1, and Supplementary Fig. 2.1.

No break in hospital referrals was associated with the March 18 to 25, 2020 triage protocols. The first break for hospital referrals occurred on February 7, 2020 (Additional File 2, Supplementary Fig. 2.1a), a week after the enactment of norms on hospitalization of COVID-19 cases by the DGSPCM (Additional File 1). Other significant breaks occurred after the first hospitalization for COVID-19 on February 25, 2020.

Three event breaks preceded the LTCF lockdown on March 8, 2020: 1) Break for hospital referrals of LTCF residents (Additional File 2, Supplementary Fig. 2.1a), 2) Break for daily deaths in persons aged 65 + living in the CoM (Additional File 2, Supplementary Fig. 2.1b-i), 3). Break for the total number of in-LTCF deaths (Additional File 2, Supplementary Fig. 2.1c-i). No break was associated with daily hospital deaths (Additional File 2, Supplementary Fig. 2.1d-i).

Figure 3c and Additional File 2, Supplementary Fig. 2.1e (hospital-to-LTCF ratios of daily deaths) summarize the results of analyses on the breaks. The regime of low ratios lasted from March 8, 2020 to April 13, 2020. On April 7, 2020 the GCM Minister of Health announced in a radio interview the gradual relaxing of restrictions on access to hospital care for LTCF residents (Additional File 1). According to the break test for the hospital-to-LTCF ratios of daily deaths, the March 8, 2020 to April 13, 2020 regime coincided with the regime of interventions of the CoM (Figs. 3c and Additional File 2, Supplementary Fig. 2.1e) with a lag of five days. The March 18–25, 2020 triage protocols were encompassed within that period.

(3) Multivariate time-series analyses of the association of LTCF residents' daily hospital referrals and daily in-hospital and in-LTCF deaths with the March 18–25, 2020 triage protocols, using a multivariable data-generating model.

Full analyses of the multivariate model (MGARCH(1,1)) are available in Additional File 3, Supplementary Table 3.1 and Supplementary Figs. 3.1, 3.2, and 3.3. Therein, the focus is on the March-April 2020 period of the COVID-19 epidemic and is limited to the graphical representation of the estimated associations between responses and impulses in the three-equation model (Fig. 2). Figure 2

Contribution of breaks and regimes to responses

In the multivariate model, no break or regime was associated with the enactment dates of the triage protocols. Within the March-April 2020 period of the COVID-19 epidemic, only the March 8, 2020 to April 13, 2020 regime from the hospital-to-LTCF ratios of daily deaths was associated with the responses (Additional File 3, Supplementary Table 3.1).

The contributions of the March 8, 2020 to April 13, 2020 regime are given by the regression coefficients in the MGARCH(1,1) model B (Additional File 3, Supplementary Table 3.1). The value of the coefficient associated with the regime is set at -15.351 for daily hospital referrals (Table Supplementary 3.1, Model B Eq. 1). Thus, the regime contributed to a loss of 15 daily hospital referrals for a total of 568 hospital referrals from March 8, 2020 to April 13 a reduction of 17% over the regime.

The estimated regression coefficients for daily in-hospital and in-LTCF deaths contribution of the March 8-April 13, 2020 regime to in-hospital and in-LTCF deaths were − 2.474 and 3.846, respectively (Additional File 3, Supplementary Table 3.1, Model B Eq. 2,3). The regime contributed a reduction of 2.5 in-hospital deaths daily and an increase of 3.9 in-LTCF deaths daily, a total reduction of 92 (6.7%) in-hospital deaths and a total increase of 142 in-LTCF deaths (2.4%). Finally, a shift contributed to a reduction of 11 in-LTCF deaths on March 8, 2020 (Additional File 3, Supplementary Table 3.1, Model B Eq. 2,3).

Contribution of impulse time series to response time series

Two statistically significant groups of regimes characterized impulses and responses during the COVID-19 epidemic period, except for daily in-hospital deaths for which no break was identified (Additional File 2, Supplementary Fig. 2.1d-i). Each of these two regimes was delimited by one break date located at the March 27, 2020 and April 2, 2020 peaks in the time series (Supplementary Figs. 2.1a-i, b-i, c-i). The first group of regimes was associated mostly with the ascending phase of daily deaths in the population aged 65+, and in LTCFs and hospitals. These regimes coincided with the GCM intervention period within a week at the beginning and end of the regimes. In the second group, regimes were linked to the descending phase of the daily death time series. Regimes were considered in the graphical analyses of the associations of responses with impulses obtained from the MGARCH(1,1) (Additional File 3, Supplementary Table 3.1, Model B)

The first foundational impulse in the data-generating model (Fig. 2) is the time series for the daily deaths in the population aged 65 + in the CoM. The contributions of this impulse to the daily in-hospital deaths of LTCF residents (blue) and LTCFs (red) are shown in Fig. 4a through solid lines for the first regime (March 1–26, 2020) and dashed lines for the second regime (March 27, 2020 -April 26, 2020). They followed bell-shaped curves over the March-April 2020 COVID-19 period and beyond in the post-COVID period. At the peak of the death curves, the estimated number of deaths in the CoM population aged 65 + contributed to 34.3% of all-cause in-LTCF deaths (calculation shown on request) and 69.0% of in-hospital deaths of LTCF residents.

In Fig. 4b, LTCF residents' deaths in-hospital (green) and in-LTCF (red) are responses to hospital referrals (blue), while hospital referrals are responses to deaths in the population aged 65+. In the first regime, the contributions to hospital referrals of daily deaths in the population aged 65 + were null or negative, while they followed an approximate bell-shaped curve in the second regime. Hospital referrals decreased in the ascending phase of the CoM death curve for the population aged 65 + and increased during the peak period of the curve. Thereafter, they decreased at a lower rate than daily deaths in the population (Fig. 4b).

The contributions of hospital referrals to in-hospital daily deaths (green) and in-LTCF deaths (red) were negative in the first regime, whereas in the second regime, their contributions were positive.

Insert Fig. 4 here

Figure 4. Contribution of impulses to responses in the MGARCH(1,1) model

a. Contribution of the March 8, 2020 to April 13, 2020 regime

- b. Contribution of impulses to hospital referrals, in-hospital and in-LTCF daily deaths of LTCF residents

The GCM interventions coincided with the ascending phase, while their gradual retreat was implemented at the beginning of the descending phase – on April 7, 2020. To summarize the results, in the ascending phase of the COVID-19 curve, the GCM interventions blocked the association of responses with impulses expected in the data generation model in Fig. 2. The associations became positive only in the descending phase (Fig. 4b).

Volatilities in responses and adjustment of predicted to observed responses

Volatilities associated with the COVID-19 epidemic were obtained only for the predicted daily in-LTCF deaths (Additional File 3, Supplementary Fig. 3.4c-ii). Predicted hospital referrals did not show statistically significant volatilities, while the pattern of volatilities associated with daily in-hospital deaths of LTCF residents occurred over the whole January 5, 2020 to June 27, 2020 period. Thus, evidence for turbulence in the March-April 2020 COVID-19 epidemic period was available only for daily in-LTCF deaths.

Summary of results

The main results of the analyses are summarized into the following five points:

While hospital referrals from LTCFs were not associated with the enactment dates of the triage protocols (March 18–25), they were associated with the GCM intervention period (March 3 to April 7). .

In-LTCF deaths were associated with deaths in the CoM population aged 65+.

LTCF daily hospital referrals and daily in-hospital deaths decreased in the GCM intervention period compared to the pre-COVID period, while daily in-LTCF deaths increased.

Trends in LTCF residents' daily hospital referrals, and daily in-LTCF and in-hospital deaths changed with the ascending and descending phases of daily deaths in the CoM population aged 65 + such that:

Hospital referrals and in-hospital deaths of LTCF residents decreased in the ascending phase of the daily deaths time series;

Past the break date of the hospital referrals time series (April 1), daily in-hospital deaths increased with daily hospital referrals.

Past the break date of the daily in-LTCF deaths (April 2, 2020) daily in-hospital deaths increased with daily in-LTCF deaths.

Evidence for turbulence was found only in the daily in-LTCF deaths time series. There was no evidence of turbulence in hospital referrals and daily in-hospital deaths of LTCF residents associated with the GCM interventions.

Interpretation of results

During the March-April 2020 peak period of the COVID-19 epidemic, the CoM’s hospitals were confronted with bed saturation and, in some cases, near collapse due to high levels of non-ICU and ICU occupancy [29, 34]. In this context, our results suggest the GCM interventions prevented hospitals from providing care to LTCF residents in the ascending phase. As indicated by the results on volatility, LTCFs were hard hit with the challenges of providing health and social care to severely ill and dying residents in times of high turbulence, while hospitals were saved from instability by decreased hospital referrals of LTCF residents in the COVID-19 epidemic.

Hospital resources were scarce for non-COVID-19 patients in the March-April 2020 peak period of the COVID-19 epidemic. Condes & Arribas[34] and Sanchez-Ubeda et al.[29] documented the overflow of ICU and non-ICU hospital beds at the end of March 2020. Sanchez-Ubeda et al. [29]included all hospital beds available from March 18, 2020 to May 27, 2020. All acute care hospital beds in the CoM were occupied from March 27, 2020 to April 5 while the ICU bed occupancy rate was over 250% in the same period [29]. Meanwhile, COVID-19 patients occupied 90% of ICU beds and two-thirds of non-ICU beds [29]. Condes & Arribas[34] included a subset of available hospital beds from March 8, 2020 onwards. By this date, deaths in the CoM population aged 65 + were increasing, and hospital referrals of LTCF residents decreased. These events occurred in synchrony with the increasing occupancy of hospital ICU and non-ICU beds by COVID-19 patients (Fig. 5), and with the start of the GCM’s Ministry of Heath’s Gerencia de Hospitales’ (Hospital Management Group). This group was responsible for the management of hospital beds and the implementation of the GCM’s Plan de Elasticidad de Recursos en Atencion en Hospitales (Hospital Resources Elasticity Plan) [18](page 45). Unexpectedly, after the enactment of the fourth version of the triage protocol on March 25 non-ICU and ICU bed occupancy increased up to their highest point from March 27, 2020 to April 4, 2020. Thus, the CoM's pre-triage protocol interventions reduced the pressure on hospitalizations by decreasing hospital referrals in the ascending phase of the March-April 2020 COVID-19 epidemic, while post-triage hospital referrals decreased with decreasing hospital bed occupancies in the descending phase (Fig. 5). The same conclusion is reached using the Sanchez-Ubeda et al.[29] daily ICU and non-ICU hospital bed availability data.

Insert Fig. 5 here

Figure 5. Hospital bed occupancy, deaths in the population aged 65+,

and hospital referrals of LTCF residents.

The data-generating model in Fig. 2 was based on two foundational impulses: the March-April 2020 period of the COVID-19 epidemic in the CoM and the March 18–25, 2020 triage protocols. This study focuses on protocols central to the public debate on CoM policies during the COVID-19 epidemic. The results of the study showed that the triage protocols were not associated with further decreases in hospital referrals. They increased after March 23, 2020 and their negative associations with the CoM daily deaths in the population aged 65 + stopped, changing from negative to positive thereafter (Fig. 4b). Hospital referrals began to decrease on March 6 coinciding with increasing daily deaths in the CoM population aged 65+ (March 1, 2020) and with increasing ICU and non-ICU hospital bed occupancy. These results suggest modifications of the data-generating model shown in Fig. 2: 1) The second foundational impulse, triage protocols, is deleted. 2) The availability of ICU and non-ICU hospital beds is introduced as the second foundational impulse. 3) The third foundational impulse is the set of interventions introduced by the GCM between March 8, 2020 and April 7, 2020. The GCM interventions changed the interplay between the impulses and responses and radically modified LTCF residents' access to hospital care. 4) Hospitals were protected against turbulence, i.e. unpredictability, in terms of the range of daily variation of deaths of LTCF residents. Turbulence associated with the daily deaths in the March-April 2020 period of the COVID-19 epidemic among LTCF residents was concentrated in LTCFs.

Limitations and strengths

There are no official statistics on the deaths of LTCF residents in Spain [35]. During the March-April 2020 epidemic period, the CoM Ministry of Social Policies, Families, Equality and Fertility collected information on the deaths of LTCF residents by place of death. Prediction equations to obtain the estimated pre-COVID-19 LTCF resident in-hospital daily deaths were modeled using the data-generating model (Fig. 2). Therefore, LTCF residents' daily in-hospital deaths during the pre-COVID-19 period were imputed based on hospital referrals, in-LTCF total daily deaths, and the CoM population aged 65 + in the post-COVID-19 period. The estimated pre-COVID-19 daily in-hospital deaths time series reproduced the pattern of associations between the post-COVID daily in-hospital and in-LTCF deaths.

Restrictions on hospital referrals of LTCF residents did not apply to those who had private health insurance [36]. An unknown number of LTCF residents with private health insurance during the epidemic had access to private ambulance services and referral to private hospitals. However, the number of LTCF residents admitted to private hospitals in the March-April 2020 period of the COVID-19 epidemic is known to be small [37]. Finally, health policy restrictions were applied only to public hospitals, not to private hospitals [37].

The data-generated model (Fig. 2) focused on the objectives of the CoM triage protocols and their implementation in the context of the March-April 2020 period of the COVID-19 epidemic. In this context, expected coefficients and their signs were hypothesized. Statistical tests on estimated coefficients examined the extent of their alignment with the data-generated model. Causality was not inferred from the results of the analyses.

This study has several strengths. It is based on official statistics of hospital referrals and deaths and uses advanced and appropriate statistical methods. All LTCF resident deaths between January 5, 2020 and June 27, 2020 were included in this study. Univariate statistical tests were used to identify breaks associated with the triage protocols. Before testing breaks, time series were examined. Breaks were tested on appropriately transformed time series. The multivariate analysis was based on a data-generating model. The MGARCH(1,1) procedure was selected based on the results of volatilities of the univariate analyses. The results of the MGARCH(1.1) were reliable based on tests on residuals and unit roots. Statistical tests and estimated MGARCH(1,1) coefficients obtained from the statistical methods used in this paper were interpreted using graphical representations.

Three conclusions can be drawn from this research:

1) Results of the multivariate time series analyses provide an evidence-based description of the associations between hospital referrals and in-LTCF and in-hospital mortality of LTCF residents in the March-April 2020 period of the COVID-19 epidemic in the CoM.

2) Decreases in LTCF residents' hospital referrals and the stability of in-hospital mortality preceded the enactment of the triage protocol by two weeks. Thus, the data-generated model (Fig. 2), driven by the objectives and means of the triage protocols, was rejected.

3) Two impulses ought to be included in a modified data-generated model: a) hospital UCI and non-UCI bed occupancies; and b) GCM policy interventions restricting hospital care for LTCF residents from March 3, 2020 to April 28, 2020.

The situation of the CoM’s LTCFs was the epitome of point 21 of the European Parliament resolution on the COVID-19 pandemic [38], which stressed that EU member states focused on preserving hospital capacity while neglecting LTCFs, resulting in excessive mortality.

AC: Autonomous Community

CoM: Community of Madrid

DGSPCM: Dirección General de Salud Pública de la Comunidad de Madrid (CoM’s Directorate General of Public Health)

GCM: Government of the Community of Madrid

ICU: Intensive Care Unit

LTCF: Long Term Care Facility

OECD: Organization for Economic Co-operation and Development

Ethics approval and consent to participate:

Not applicable. All data are from administrative files aggregated at the level of facilities on a daily basis.

Consent for publication :

Not applicable.

Availability of data and materials :

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Competing interests :

The authors declare that they have no competing interests.

Funding :

No funding were available for this study

Authors' contributions :

FB contributed to the definition of the objectives of the study, developed the study’s data-generating model, designed the analytical and statistical models, and proceeded to the statistical analyses. He contributed to the interpretation of results, and to the discussion and conclusion sections of the paper. He was a major contributor in writing the manuscript. MVZ contributed to the definition of the objectives of the study, to data analyses, to their interpretation, and to the writing of the paper. MVZ and FJGL searched for, obtained and formatted the administrative data files. FJGL contributed to the definition of the objectives of the study. FJGL and FPR revised the manuscript, suggested modifications, where contributors to the analyses and interpretation of the results.

Acknowledgements

The authors thank Ms. Claudia Loughran for her editorship.

Sapiano MRP, Dudeck MA, Soe M, Edwards JR, O’Leary EN, Wu H, et al. Impact of coronavirus disease 2019 (COVID-19) on US Hospitals and Patients, April-July 2020. Infect Control Hosp Epidemiol. 2022;43:32–9.
Mateen BA, Wilde H, Dennis JM, Duncan A, Thomas N, McGovern A et al. Hospital bed capacity and usage across secondary healthcare providers in England during the first wave of the COVID-19 pandemic: A descriptive analysis. BMJ Open. 2021;11.
Verelst F, Kuylen E, Beutels P. Indications for healthcare surge capacity in European countries facing an exponential increase in coronavirus disease (COVID-19) cases, March 2020. Eurosurveillance. 2020;25.
Sepulveda ER, Stall NM, Sinha SK. A Comparison of COVID-19 Mortality Rates Among Long-Term Care Residents in 12 OECD Countries. J Am Med Dir Assoc. 2020;21:1572–1574e3.
Aalto UL, Pitkälä KH, Andersen-Ranberg K, Bonin-Guillaume S, Cruz-Jentoft AJ, Eriksdotter M, et al. COVID-19 pandemic and mortality in nursing homes across USA and Europe up to October 2021. Eur Geriatr Med. 2022;13:705–9.
Islam N, García López FJ, Jdanov DA, Ángel Royo-Bordonada M, Khunti K, Lewington S et al. Unequal impact of the Covid-19 pandemic on excess deaths, life expectancy, and premature mortality across Spanish regions in 2020 and 2021. https://doi.org/10.1101/2021.11.29.21266617.
Martinez-Peromingo J, Serra-Rexach JA. Long-Term Care Facilities and the COVID-19 Pandemic: Lessons Learned in Madrid. J Am Geriatric Soc. 2020;68:1920–2.
Zunzunegui MV, Rico M, Béland F, García-López FJ. The Impact of Long-Term Care Home Ownership and Administration Type on All-Cause Mortality from March to April 2020 in Madrid, Spain. Epidemiologia. 2022;3:323–36.
Koleva G, Rico M, García López FJ, Figuera D, Padilla J, García M. The impact of COVID-19 in nursing homes in Madrid, Spain: a need for assessment. Lancet Reg Health - Europe. 2021;11.
Zalakaín J, Davey V, Zalakain J. The COVID-19 on users of Long-Term Care services in Spain 1. 2020.
Comas-Herrera A, Zalakaín J, Lemmon E, Henderson D, Litwin C, Hsu AT et al. Mortality associated with COVID-19 in care homes: international evidence. 2020.
Zunzunegui MV, Béland F, Rico M, López FJG. Long-Term Care Home Size Association with COVID-19 Infection and Mortality in Catalonia in March and April 2020. Epidemiologia. 2022;3:369–90.
Zunzunegui MV, Béland F, García López FJ. Restrictions on Hospital Referrals from Long-Term Care Homes in Madrid and COVID-19 Mortality from March to June 2020: A Systematic Review of Studies Conducted in Spain. Epidemiologia. 2023;4:176–87.
Reyero Zubiri A. Morirán de forma indigna. K.O. ediciones. Madrid; 2022.
Rico M. Los seis documentos que demuestran que Ayuso miente sobre la orden de no trasladar enfermos de residencias a hospitales. Infolibre. 2020.
Dirección General de Coordinación Sociosanitaria. Consejería de Salud. Comunidad de Madrid. Protocolo de coordinación - Madrid -Covid- 25/03/2020. 2020:1–7. https://amyts.es/wp-content/uploads/2020/01/COVID-CAM-coordinaci%C3%B3n-residencias-sinfecha-20200412.pdf. Accessed 19 Oct 2023.
Dirección General de Coordinación Sociosanitaria. Consejería de Sanidad de la Comunidad de Madrid. Protocolo de coordinación - Madrid - Covid – 18/03/2020. 2020:1–5. https://files.mediaset.es/file/10002/2020/05/19/PROTOCOLO_DE_COORD-_PARA_LA_ATENCION_DE_PACIENTES-PDF_-2_2b93.pdf. Accessed 19 Oct 2023.
Government of the Community of Madrid. Memoria 2020. Servicio Madrileño de Salud. 2020. https://gestiona3.madrid.org/bvirtual/BVCM050404.pdf. Accessed 23 Sep 2023.
Zhang XS, Charland K, Quach C, Nguyen QD, Zinszer K. Institutional, therapeutic, and individual factors associated with 30-day mortality after COVID-19 diagnosis in Canadian long-term care facilities. J Am Geriatr Soc. 2022. https://doi.org/10.1111/jgs.17975.
Costa-Font J, Jiménez Martin S, Viola A. Fatal Underfunding? Explaining COVID-19 Mortality in Spanish Nursing Homes. J Aging Health. 2021;33:607–17.
Sergi JC-F, Cristina J, Prieto V, Costa-Font J. Long-term Care in Spain. 2022.
Boletin Oficial del Estado. Orden SND/265/2020, de 19 de marzo, de adopción de medidas relativas a las residencias de personas mayores y centros socio-sanitarios, ante la situación de crisis sanitaria ocasionada por el COVID-19. Boletin Oficial del Estado. 2020. https://www.boe.es/eli/es/o/2020/03/19/snd265/con. Accessed 23 Sep 2023.
Consejo de MInistros. Plan de desescalada. 2020. https://www.lamoncloa.gob.es/consejodeministros/Paginas/enlaces/280420-enlace-desescalada.aspx. Accessed 4 Nov 2023.
Ditzen J, Karavias Y, Westerlund J. Testing and Estimating Structural Breaks in Time Series and Panel Data in Stata. 2021.
Rapach DE, Strauss JK. Structural breaks and GARCH models of exchange rate volatility. J Appl Econom. 2008;23:65–90.
Schaffer AL, Dobbins TA, Pearson SA. Interrupted time series analysis using autoregressive integrated moving average (ARIMA) models: a guide for evaluating large-scale health interventions. BMC Med Res Methodol. 2021;21.
Becketti S. Introduction to Time Series Using Stata. Revised edition. 2020.
Stata. Stata Press. 2017.
Sánchez-Úbeda EF, Sánchez-Martín P, Torrego-Ellacuría M, Rey-Mejías Á, Del, Morales-Contreras MF, Puerta JL. Flexibility and bed margins of the community of madrid’s hospitals during the first wave of the sars-cov-2 pandemic. Int J Environ Res Public Health. 2021;18.
Procedimiento ordinario 434. /2020 P-01 Ayuntamiento de Leganés 6 de Mayo 2020. Auto Tribunal Superior de Justicia de Madrid; 2020.
Amnesty International. ABANDONADAS A SU SUERTE - La desprotección y discriminación de las personas mayores en residencias durante la pandemia COVID-19 en España. 2020.
Fonseca-Herrero Raimundo AJ. Auto Tribunal Supremo de Justicia Sala de lo Contencioso. Procedimiento 7770/2022. 2023.
Galaup L, Caballero F. Cronología del desastre en las residencias de Madrid: 7.600 fallecidos, crisis en el Gobierno e investigación judicial. El diario. 2020.
Condes E, Arribas JR. Impact of COVID-19 on Madrid hospital system. Enferm Infecc Microbiol Clin. 2021;39:256–7.
Zunzunegui MV, García López FJ, Rodríguez V. Unknown mortality of the population in long-term care homes in Spain. Gac Sanit. 2023;37.
Peinado F. Los mayores con seguro privado sí fueron trasladados de residencias a hospitales en Madrid. El País. 2020.
Peinado F. Un empresario de hospitales y otro de residencias desmontan la defensa de Ayuso en la crisis de los geriátricos. El País. 2020.
Montserrat MD, European Parliament. P9_TA(2023)0282 COVID-19 pandemic: lessons learned and recommendations for the future. 2023.

Tables 1-2 is available in the Supplementary Files section.

No competing interests reported.

Download PDF

Journal Publication

published 14 Aug, 2024

Read the published version in BMC Geriatrics →

Editorial decision: Revision requested
13 Feb, 2024
Reviews received at journal
10 Jan, 2024
Reviewers agreed at journal
31 Dec, 2023
Reviewers invited by journal
31 Dec, 2023
Editor assigned by journal
31 Dec, 2023
Editor invited by journal
30 Dec, 2023
Submission checks completed at journal
30 Dec, 2023
First submitted to journal
22 Dec, 2023

You are reading this latest preprint version

Hospital referrals, exclusions from hospital care, and deaths among long-term care residents in the Community of Madrid during the March-April 2020 COVID-19 epidemic period: A multivariate time series analysis

Status:

Journal Publication

Version 1

Abstract

Background:

Methods:

Results:

Conclusions:

Figures

Background

Methods

Material

Statistical procedures

Data visualization

Univariate analyses of changes in hospital referrals and daily mortality

The case of hospital referrals

Multivariate analyses of hospital referrals and in-LTCF and in-hospital deaths

Results

Contribution of breaks and regimes to responses

Contribution of impulse time series to response time series

Volatilities in responses and adjustment of predicted to observed responses

Discussion

Summary of results

Interpretation of results

Limitations and strengths

Conclusion

Abbreviations

Declarations

References

Tables

Additional Declarations

Supplementary Files

Status:

Journal Publication

Version 1