Study design to determine the effects of widespread restrictions on hospital utilization to control an outbreak of SARS in Toronto, Canada

Abstract

CONTEXT: Efforts to control an outbreak of severe acute respiratory syndrome (SARS) in Toronto, Canada, led to the imposition of major restrictions on nonurgent use of hospital-based services. OBJECTIVE: To describe a methodology to determine the impact of the restrictions on healthcare utilization. DESIGN, SETTING, POPULATION: Population-based study of the Greater Toronto area and unaffected comparator regions, before, during and after the SARS outbreak (April 2001 to March 2004). OUTCOME MEASURES: Population-based rates of hospital admissions, emergency department and outpatient visits, inter-hospital transfers, diagnostic testing and essential drug prescribing, adjusted for age and sex. METHODS: We will assess the temporal patterns of healthcare utilization rates before, during and after the SARS restrictions in different regions using administrative health databases and longitudinal data analysis methods (generalized estimating equations). We will also use longitudinal cohort models to assess the effects of the restrictions to outcomes in cohorts diagnosed with specific chronic diseases. CONCLUSION: We will document the extent to which utilization of healthcare services decreased during the SARS epidemic and identify clinical problem areas where SARS-related restrictions had adverse consequences on health. This work will have planning implications for future outbreaks of SARS or other infectious diseases. Understanding how the outbreak control measures affected use of health services and ultimately the health of the population, is an important part of understanding the impact of SARS restrictions.

Keywords:

• influenza
• methodology
• outbreak control
• pandemic
• SARS

The outbreak of severe acute respiratory syndrome (SARS) in the Greater Toronto Area (GTA) was an event without precedent in Canada since the advent of universal healthcare. It led not only to direct morbidity and mortality related to SARS, but also to severe and widespread restrictions of the GTA’s healthcare system in an effort to control the disease. Although these restrictions appear to have been effective in controlling the wider spread of the outbreak [1], they may have produced unintended health consequences for the non-SARS population [2]. The literature on health emergencies, disasters, bioterrorism and civil defense planning has largely focused on the surge in care required to accommodate those directly affected by the disaster [101,102]. Implicit in these scenarios is a shift of attention away from pre-existing population needs and an assumption that pre-existing population demand for health services will remain constant during the crisis. Little consideration is given to the possibility that the delivery of health services to meet pre-existing demands may not remain constant during an emergency, either due to reduced access to healthcare services and/or changes in population healthcare-seeking behavior.

Our underlying hypothesis is that the dramatic measures taken to control the SARS emergency may have had unintended adverse consequences on access to care and maintenance of health and health services for the rest of the GTA population. Our purpose is to outline a methodology to identify clinical areas in which reduced access to medical care, combined with reduced demand for health owing to fear of contracting SARS among the population, resulted in, first, declines in appropriate healthcare utilization and second, potentially harmful consequences for patient and population health.

Description & chronology of the SARS restrictions

The GTA’s index SARS case arrived in Toronto on February 23, 2003. The outbreak would eventually become the largest outside of Asia, with 257 probable cases and 43 deaths [103]. The outbreak was largely confined to hospitals, with healthcare workers accounting for more than half of all SARS cases, including three of the deaths. Fears that the outbreak could worsen within hospitals or spread into the community led to the declaration of a Provincial Health Emergency and the imposition of wide-ranging restrictions on health services. These restrictions applied to all 32 hospitals in the GTA since the extent of the spread of SARS within hospitals was unclear. The directives restricted ambulatory and in-patient medical and surgical activity to urgent cases only, severely restricted visitors, expanded respiratory isolation rooms and mandated the use of personal protective equipment by staff in high-risk areas. Given the large number of tertiary care centers within the GTA, this meant reduced access to highly specialized services, such as cardiac catheterization and MRI, specialized programs, such as oncology, neurovascular diseases and trauma, and difficult access to referral for off-site specialty care, both within the GTA and in surrounding communities that depend on the GTA for highly specialized care. Three community hospitals where unprotected SARS transmission took place were closed to new admissions, all surgery was halted and the emergency departments and outpatient clinics were closed for periods ranging from a few weeks to several months. A centralized system was created for real-time screening of all inter-hospital patient transfers in Ontario, since the transfer of unrecognized SARS cases had allowed the outbreak to spread between hospitals [3]. The restrictions were proclaimed in a blanket fashion without specific measures designed to mitigate the impact on potentially vulnerable subgroups of patients, such as those with chronic disease, the elderly or the poor. The effects of these restrictions were compounded by patient fear of seeking care in hospitals or doctors’ offices, given the well-publicized risk of acquiring SARS in healthcare settings.

Conceptual framework of the effects of the SARS restrictions

These changes to the healthcare system may have had unintended consequences. Medical care epidemiologists have previously studied the associations between health system resources, utilization and outcomes in healthcare systems with very different levels of resources. Previous research has demonstrated that the magnitude of variations in hospitalization rates for medical and surgical conditions depends on many factors besides illness, such as health system capacity (e.g., hospital beds and physicians per capita), physician practice style, medical consensus on appropriate treatments and patient healthcare seeking behavior [4–8]. For some conditions, known as high variation conditions (e.g., chronic obstructive pulmonary disease [COPD] and diabetes), regional variations in admission rates are striking, while for others, known as low variation conditions (e.g., acute myocardial infarction [AMI] or hip fracture), there is relatively little variation [5,6]. Low variation conditions are those where admission rates appear to reflect population illness rates since patients always seek care, there is little difficulty making the diagnosis and hospital admission is mandatory; high variation conditions are those where admission rates reflect illness as well as nonhealth factors, such as local health system capacity.

Intermittent variations in healthcare utilization, outcomes and quality of care have also been found within regions. Studies of medical staff strikes suggest that they are associated with changes in utilization of healthcare services [9–11], a decrease in mortality due to reduced surgical activity [10,12], and poorer quality of care [13–16], especially for patients of lower socioeconomic classes [17]. A study that compared patients hospitalized on weekends and weekdays across Ontario also found that in-hospital mortality was higher for some conditions, such as pulmonary embolus (PE) and ruptured abdominal aortic aneurysm (AAA), but not others, such as AMI and hip fracture [18]. Mortality was higher among conditions where regional variations are more pronounced and more physician discretion exists in management [4,6] or where more complex coordination of the healthcare system, such as inter-hospital transfers, is required [18].

However, these studies may only be partly generalizable to SARS-related capacity restrictions for a number of reasons. First, studies of patient outcomes in high versus low resource settings compare the performance of systems in steady state, where resource levels are static or changing gradually and the resources available in ‘low’ resource settings are still adequate by virtually any standard (e.g., the prototypical low resource US region was New Haven, CT [7]). Similarly, the impact of strikes may not be relevant because they are anticipated, explicitly exclude high-risk patient groups (e.g., emergency department [ED] patients or urgent surgery) and are usually short-lived. Finally, research on day-of-the-week variation is not directly generalizable since weekends recur, are brief, have staff and resources planned accordingly (i.e., some healthcare resources [e.g., EDs] function with the same or increased staff on weekends) and additional resources are available at short notice if required (i.e., on-call schedules). A final difference between all these situations and the SARS outbreak was the degree of patient anxiety and fear about the risk of exposure to a new and often fatal illness when using the healthcare system, which may have substantially altered patients’ healthcare seeking behavior. Nonetheless, these studies suggest that acute changes to a healthcare system can be associated with adverse outcomes, but only for certain conditions.

During the period of SARS-related restrictions, we hypothesize that, first, differences in utilization or outcomes did not occur in conditions where there is strong medical consensus on patient management, where little variation in patient healthcare seeking behavior exists and where the healthcare system capacity was sufficient and its coordination straightforward, despite the SARS restrictions (low variation conditions). Examples include AMI, hip fractures and births [5,6]. Patients with these conditions often have symptoms that are difficult to ignore (severe chest pain, inability to walk or labor pains) and are therefore not likely to delay presentation to hospital. Management of these conditions is subject to strong medical consensus, such as the need for hospitalization, aspirin and β-blocker therapy for AMI patients. These patients typically do not require complex coordination of health system resources and, in most hospitals in the GTA, the acute phase can be managed without the need for inter-facility transfer. Since these constitute only approximately 10% of admissions [6], the magnitude of capacity restrictions observed in the SARS epidemic would not be expected to be sufficiently severe to affect admission rates for these patients.

Second, we hypothesize that acute reductions in health services affect outcomes for conditions where there is substantial physician discretion in medical decision making regarding treatment and disposition, which requires complex coordination of the healthcare system (including inter-hospital transfers to access tertiary care), or where substantial portions of care are delivered outside of critical care units, and which are frequently fatal if not adequately treated (higher variation conditions). Examples of such acute conditions are PE, intracranial hemorrhage, gastrointestinal (GI) bleeding and cancers of the respiratory tract. Care of patients with these conditions is complex and often requires transfer to sub-specialist care. These may present with common symptoms more easily ascribed to benign causes (e.g., mild shortness of breath, headache, back pain) or exacerbations of pre-existing symptoms in cancer patients (e.g., pain, weakness, fever). These symptoms allow for greater discretion in physician management and for greater variations in patient healthcare-seeking behavior. During SARS, both may have been altered either owing to more difficult access to diagnostic tests or transfer to subspecialist care and/or because patients, anxious to avoid contact with the healthcare system, may have delayed presentation to hospital.

Ambulatory and primary care services also probably experienced substantial reductions in utilization during the period of SARS-related restrictions. Some patient populations may be vulnerable to even slight reductions or disruptions in routine care. Ambulatory care sensitive conditions (ACSC) are defined as those for which lack of access to timely and effective outpatient care may result in ‘avoidable’ or ‘preventable’ hospitalization [19,20]. Examples are diabetes, congestive heart failure (CHF), COPD and asthma. Hospitalizations for ACSCs may reflect reduced access to care rather than prevalence of disease, patient propensity to seek care or physician practice style [21,22]. For example, a recent study in Ontario has shown that hospitalizations for complications of diabetes, such as hyperglycemia and ketoacidosis, have decreased in recent years, perhaps owing to improved access to primary care [23]. Access restrictions may reverse this. Other care, such as adult annual health examinations, which include evidence-based elements such as fecal occult blood test (FOBT) and pap smears, may be considered discretionary or low priority during acute restrictions [24,25]. Immunization and essential drug utilization for chronic illnesses that are largely asymptomatic (e.g., hypertension or hyperlipidemia) [26] may be similarly affected, although disruptions in immunization may be ill-advised during many emergencies and discontinuity of essential drugs may lead to complications. We expect reduced utilization of evidence-based primary care thought to be deferrable during the SARS period.

Specific study aims

The study we propose would have, as its main objective, the description of the patterns of healthcare utilization before, during and after the SARS outbreak in the GTA, circum-GTA, Ottawa and London populations. We will examine primary and ambulatory care, out-patient visits, evidence-based drug prescriptions, emergency department visits, and hospital admissions for ACSC, hypothesizing a reduction in utilization during SARS and a return to expected levels in the post-SARS period. Second, we will assess acute conditions requiring hospitalization where medical consensus exists regarding diagnosis and the need for hospitalization. Here, we hypothesize there will be no difference in hospital admissions or outcomes during SARS. Third, we will assess acute conditions requiring hospitalization where substantial discretion exists in the diagnosis or management or which require complex coordination of multiple health system resources. Here, we hypothesize no difference in hospital admissions, but an increase in mortality during SARS. In addition, we will assess whether vulnerable subgroups were affected differentially, specifically, the elderly and the poor.

Research design & methods

The analyses of population rates will provide information about general trends but are essentially ecological in nature. We will therefore perform a series of longitudinal cohort analyses on potentially vulnerable subgroups (ACSCs and acute hospitalized conditions), where we postulate specific causal mechanisms through which the restrictions acted to change the pattern of healthcare delivery and, therefore, may have caused potential harm.

Data sources

Virtually all data utilized in this study reside in health administrative databases permanently stored at the Institute for Clinical Evaluative Sciences (ICES) for research purposes. These databases are regularly updated at intervals varying from monthly to annually. The data are subject to stringent quality control verifications prior to their transfer to ICES to ensure validity. All databases contain unique patient identifiers to enable their linkage; however, they are encrypted to protect privacy.

Study regions

The four study regions are the GTA, the circum-GTA, Ottawa and London. Regions were defined as collections of contiguous Hospital Service Areas (HSAs), representing local healthcare markets for community-based in-patient care, using previous methods [27,28]. The GTA region contains all HSAs of hospitals subject to SARS restrictions. Ottawa and London are ideal control regions because they are large urban centers having multiple hospital sites, including several tertiary hospitals, and they form organizationally complex, independent, self-contained healthcare systems. They are also sufficiently geographically distant that GTA patients are unlikely to seek medical care there. The circum-GTA region was created for analyses of inter-hospital transfers, defined as contiguous HSAs surrounding the GTA where routine hospitalizations tend to occur at local hospitals, but from which more than 20% of acute-care patient transfers are to GTA hospitals. This region may have been particularly vulnerable to SARS-related restrictions of inter-hospital transfers of patients. All rates are population-based and will be computed for the population residing within each region. Because of patient migration for hospital care, we will create regional boundaries using a patient origin study, as in previous work [4,28].

Description of patterns of general healthcare utilization

Patterns of healthcare utilization before, during and after the SARS restrictions in GTA, circum-GTA, Ottawa and London regions will be modeled in order to estimate the decrease in utilization within each region compared with pre-restriction levels. For each specific type of utilization (Table 1), we will begin by computing weekly use rates, starting approximately 3 years prerestrictions on April 1, 2000, through 1 year post-restrictions to March 14, 2004. These are defined as the total number of visits or admissions during a week divided by the corresponding resident population in the region, derived from census data. Adjusted rates are computed by age–sex, standardizing each region’s crude rates directly to the Ontario population [29]. This will produce age–sex-adjusted population-based rates of utilization of hospital, emergency, prescription drug and inter-hospital transfers before, during and after the restrictions, for each region. We will define the period of the SARS epidemic from March 15, 2003 (coinciding with the date of formal notification of all Ontario physicians of an outbreak of atypical pneumonia) to July 14, 2003. Mean pre-SARS rates will be computed as the weighted average of the crude weekly rates from April 1, 2000 to March 14, 2003, weighting by population. Plots of use rates would display the entire time series.

Analysis of the effects of restrictions on health services & outcomes for specific populations

From previous research, there are only a few acute conditions where admission rates are unlikely to change in the face of capacity restrictions, such as AMI, hip fracture and births [5,6,18]. Conversely, restrictions are more likely to have caused harm for conditions where substantial physician discretion exists in medical decision making regarding acute management (e.g., PE, GI bleed and respiratory cancers), which require complex coordination of the healthcare system (e.g., intracranial bleeds and respiratory cancers), where substantial portions of care are delivered outside of critical care units (e.g., PE, GI bleed and respiratory cancers) or which are frequently fatal if not adequately treated (all four conditions). For the latter, we have chosen four sentinel conditions where there is documented evidence of how harm might occur, which have moderately high incidence rates and can be validly coded using administrative data (Table 2). We anticipate that related conditions would be similarly impacted so that the research findings and policy recommendations would apply to them as well.

We will assemble separate cohorts of patients hospitalized for each condition between April 1, 2001 and November 15, 2003. Follow-up for each patient will begin on index admission date and will continue through to 6 months post-admission, censoring for mortality. We will evaluate 6-month mortality, and 60-day cause-specific and all-cause readmissions. For chronic ACSC conditions, we will focus on those examined in the first validation study [21], diabetes, congestive heart failure (CHF), COPD and asthma, since they are common, serious and can have frequent and/or severe exacerbations (Table 3). Two ACSCs, diabetes and asthma, are relevant to the pediatric population as well. The analysis will be identical to that described for acute conditions requiring hospitalization.

Analysis

To assess the patterns of healthcare utilization before, during and after the SARS restrictions in different regions, we will use longitudinal data analysis methods (generalized estimating equations [GEE]) for clustered count data [30,31]. GEE models account for correlations among repeated responses over time within the same region and allow for time-dependence of the response rate. Unlike time series models, GEE models produce consistent estimates of regression coefficients and standard errors, even when the correlation structure is mis-specified [30,32]. To assess changes in patterns of use, we will use an over-dispersed log–linear Poisson regression model, using the logarithm of the age- and sex-specific counts as dependent variable, weighting by the population in the region; all models will adjust for age group, sex and age–sex interactions. This produces estimates of relative rates from the exponentiated regression coefficients. We will include a linear or quadratic time trend and indicator variables representing calendar month, to model pre-SARS trends and seasonal fluctuations. To account for correlations among stratum-specific outcomes over time, we will cluster by age group and sex and use an auto-regressive correlation structure with a lag period of 4 weeks. Since average use rates may have differed among regions before the restrictions, even after adjusting for age and sex, each region will be treated as its own control. We will compare each region’s age–sex adjusted utilization rate for each month from March 15, 2003 through March 14, 2004 relative to the expected utilization rate based on pre-SARS restrictions patterns. The expected use rate will be computed by projecting the pre-SARS annual trend and seasonal fluctuations to each month in the SARS and post-SARS periods, adjusting for age and sex. The relative rate estimates and confidence intervals will allow us to measure the relative change in use rates during and after the restrictions, as well as the time to return to expected pre-SARS rates. The models will also be stratified by age and SES to see whether the restrictions affected different age groups (young, elderly) or SES groups differently; census data will be used to evaluate socioeconomic status at the postal code level [33].

For the analysis of health outcomes in specific patient populations, follow-up for each patient will begin on index admission date and will continue through to 6 months post-admission, censoring for mortality. For all cohorts, we will evaluate mortality and all-cause readmissions using time-to-event Cox proportional hazards models [34]; for the AMI cohort, we will also assess readmissions for recurrent AMI. All models will adjust for age, sex and comorbidities [35,36] measured during the index admission; for the AMI cohort, prior cardiac admissions and revascularization procedures will also be controlled for. Models will include a time-dependent term to indicate the pre-/peri-/post-restrictions periods defined above in order to evaluate the impacts of the restrictions.

Power

The approximate regional populations are GTA, 2.5 million; Ottawa, 800,000; and London, 400,000. For all analyses of population-based rates of services and utilization, we will have over 90% power to detect very small (1–2%) differences between the pre- and peri-SARS restrictions period; however, we anticipate much larger decreases in rates. Table 4 reports minimum detectable relative risk from the pre- to peri-restrictions period in the GTA cohorts, with 90% power using a 5% two-sided test [37], we do not anticipate significant changes in other regions’ cohorts. Power for the acute conditions is based on pooling condition-specific effects. Power will be approximately 90% to test for clinically important changes in rates of hospitalization for the ACSC cohorts and 6-month mortality for the high-variation acute condition longitudinal cohorts.

Policy importance

We believe that understanding the impact of SARS restrictions on health system utilization may provide lessons for the management of future outbreaks. Policy questions addressed by our results will include the effectiveness of restrictions in reducing elective and nonelective hospital utilization, their impact on referral patterns between hospitals in a regionalized health system, and population health-seeking behavior during major infectious disease outbreaks, as well as potentially revealing some unintended consequences of such restrictions.

Limitations

Administrative data are not collected for research purposes, so that some data inaccuracy is inevitable. However, three recent studies have confirmed the validity of coding of in-patient administrative records [38], the Ontario Drug Benefit claims data [39] and the Ontario Diabetes Database [40]. Administrative data lack the rich clinical detail found in medical charts so that full risk adjustment is not possible; however, in most of our population-based analyses, risk adjustment would not be relevant. We will be able to generally determine whether patients were more seriously ill using urgency scores for ED visit data. Decreases in utilization for preventive care, immunizations and drug prescriptions among diabetic patients and asthmatics are of public health concern, regardless of patient comorbidity. We will control for comorbidity in cohort analyses and we do not expect significant shifts in disease severity among admitted patients over the short time frame. By pre-specifying the clinical conditions of interest, we may fail to identify the adverse effects of SARS restrictions in other conditions. We chose sentinel conditions that are markers for specific mechanisms of harm documented in previous medical care epidemiology research. We anticipate that related conditions would be similarly affected so that the research findings and policy recommendations would apply to these as well.

Expert commentary

Our goals are to develop a methodology to describe the changes in healthcare utilization patterns in four Ontario regions that were affected differently by the health system restrictions during the SARS outbreak and examine the effects of the restrictions on outcomes for specific patient populations. Results from this study will enable us to draw ‘lessons learned’ that can assist healthcare decision makers to plan for future major infectious disease outbreaks and may help mitigate adverse effects resulting from infection control measures in the future. Our results and recommendations should be widely generalizable to other healthcare settings inside and outside Canada.

Five-year view

The SARS outbreak occurred at a time of heightened concern about the potential for large-scale disasters and epidemics, such as pandemic influenza. Many jurisdictions are now focusing more than ever on enhanced preparedness plans for such events. Over the next few years, planners will examine the Toronto experience as they seek to understand how similar hospital utilization restrictions might affect their own healthcare systems.

Acknowledgements

The authors would like to acknowledge the contributions of Doug Manuel, Astrid Guttmann, Merrick Zwarenstein, Andreas Laupacis and Brian Schwartz, and thank Nancy MacCallum and Lina Paolucci for their assistance in the completion of this manuscript. Michael Schull had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. The Canadian Institutes for Health Research (CIHR) funded this study. CIHR also provides career support to Michael Schull and David Alter.

Table 1. Patterns of healthcare utilization by type of utilization.

Healthcare sector	Specific types of utilization
Physician visits	Total physician visits, total primary care visits, total specialist visits (cardiologists, internists)
Emergency department visits	Total emergency department visits (overall and by triage level)
Hospital admissions	Total hospital admissions (overall, medical, surgical [elective, urgent])
Prescription drugs	Total evidence-based prescriptions filled for population 65+ years (antihyperglycemics, antihypertensives, inhalers)
Acute care inter-hospital ambulance transfers	Total inter-hospital transfers (acute care to acute care hospital, emergency department to acute care hospital)

Table 2. Health outcomes in acute conditions requiring hospitalization.

Acute conditions	Specific outcomes	Population of interest
Acute conditions where physician consensus exists regarding diagnosis and need for hospitalization	Total hospital admissions for AMI, hip fracture Total hospital admissions for births All-cause readmissions Mortality (30 day, 6 month) Readmissions for recurrent AMI	Adult population age 18+ years Adult female population age 15–44 years Cohort-specific hospitalized patients Cohort-specific hospitalized patients Patients hospitalized with AMI
Acute conditions where substantial patient/physician discretion exists in diagnosis or management or that require complex coordination of multiple health system resources	Total hospital admissions for pulmonary embolus, intracranial hemorrhage, gastrointestinal bleeds, cancer of trachea, bronchus, lung All-cause readmissions Mortality (30 day, 6 month)	Adult population age 18+ years Cohort-specific hospitalized patients Cohort-specific hospitalized patients

AMI: Acute myocardial infacrtion.

Table 3. Primary and ambulatory care health outcomes.

Healthcare service	Specific type of service/outcome	Populations of interest
Recommended preventive care	Total well-baby checks, total immunizations	Children less than 2 years old Children less than 6 years old
Essential drug prescriptions	Total antihypertensive pills dispensed, total oral hypoglycemic pills dispensed, total inhalers dispensed	Noninstitutionalized population 65+ years
Ambulatory care sensitive conditions	Hospital admissions	Population-based and cohort-specific outcomes for diabetes, severe chronic obstructive pulmonary disease, severe congestive heart failure, pediatric asthma

Table 4. Power analyses.

Parameter	Congestive heart failure	COPD	Diabetes	Pediatric asthma	Acute conditions (AMI, hip fracture)	Acute conditions (pulmonary embolism, intracranial hemorrhage, gastrointestinal bleed, respiratory cancer)
N Toronto*	5590	995	140,000	2400	20,970	6990
	Hospitalization rates				6-month mortality rates
Rate pre-restrictions	0.43	0.51	0.01	0.10	0.17	0.40
Min RR^‡	1.05	1.1	1.1	1.2	1.05	1.05

*N represents total Greater Toronto Area ambulatory care sensitive conditions cohort size on 4/1/01; and total number of acute condition admissions over 3 years.

^‡Minimum detectable relative risk.

AMI: Acute myocardial infarction; COPD: Chronic obstructive pulmonary disease.

Key issues

•	The control of Toronto’s severe acute respiratory syndrome (SARS) outbreak in 2003 required severe restrictions on the nonurgent utilization of hospitals.
•	The impact of these restrictions on the non-SARS population is unknown.
•	This study will outline methods to determine the effect of the restrictions on hospital utilization and patient outcomes.
•	These results will provide important lessons for the management of future epidemics.