Health resource use in epilepsy: Significant disparities by age, gender, and aboriginal status

Authors


Address correspondence to Dr. Nathalie Jetté, Foothills Medical Centre, 1403–29th Street NW, Calgary, Alberta, Canada T2 N 2T9. E-mail: njette@ucalgary.ca

Summary

Purpose: Epilepsy imposes a significant burden on society. The objective of this study was to estimate health resource utilization (HRU) over a 1-year period in epilepsy patients, using administrative databases.

Methods: Three administrative databases (inpatient, emergency, and physician claims) were used to identify epilepsy cases. HRU variables included general physician (GP) and emergency (ER) visits, physician billings, hospitalizations, and length of stay (LOS). Logistic regression was used to determine the association between demographic variables and HRU variations.

Results: Among the 1,431 patients with a mean age of 37.5 ± 17.3 years, 56 (4%) were aboriginal. Ninety-six percent of patients saw a GP or a specialist (outpatient visit), 12% were hospitalized, and 8% visited the ER. Younger patients were more likely to see a neurologist (OR = 1.7, 95% CI 1.3–2.3), visit the ER (OR = 4.9, 95% CI 3.2–7.4), or be hospitalized (OR = 2.9, 95% CI 2.0–4.3). Females were less likely to see a GP but more likely to see a neurologist. Aboriginals were more likely than nonaboriginals to visit the ER (OR = 2.3, 95% CI 1.1–5.0) or be hospitalized (OR = 2.8, 95% CI 1.5–5.1) but less likely to see a neurologist (OR = 0.3, 95% CI 0.2–0.6). Welfare status and residence location (urban vs. rural) were not associated with HRU level.

Discussion: We demonstrated the feasibility of using administrative databases to assess HRU in epilepsy. We also uncovered disparities in HRU by age, gender, and by aboriginal status, suggesting possible internal or external barriers to specialized care in some groups.

Various methods are used to estimate the burden of epilepsy, including health resource utilization (HRU) studies, mostly consisting of economic analyses (Diehr et al., 1999). Cost of epilepsy studies estimating the direct, and rarely the indirect cost of epilepsy, have been carried out in the United States. (Begley et al., 1994; Murray et al., 1996; Griffiths et al., 1999; Begley et al., 2000; Frost et al., 2000; Begley et al., 2002; Penberthy et al., 2005), UK (Cockerell et al., 1994; Jacoby et al., 1998; Kotsopoulos et al., 2003; Morgan & Kerr, 2004), France (De Zelicourt et al., 2000), Italy (Berto et al., 2000; Tetto et al., 2002; Beghi et al., 2004), Switzerland (Gessner & Horisberger, 1993), Sweden (Silfvenius, 1988; Persson et al., 2003), Australia (Banks et al., 1995), China (Mak et al., 1999), Africa (Nsengiyumva et al., 2004), India (Thomas et al., 2001; Krishnan et al., 2004), and in Oman (Al-Zakwani et al., 2003). A recent study comparing the cost of epilepsy across eight European countries found that the prices of medical services varied as much as 24 times, whereas the price of antiepileptic drugs varied up to 4.4 times among the countries (Heaney et al., 2001). Thus, the cost of illness analyses must be compared cautiously among different health care systems (Heaney et al., 2001). Furthermore, economic analyses often do not address variations in patterns of HRU based on geographic location or socioeconomic status that may suggest problems with access to care.

Epilepsy HRU studies, which are not economic analyses, are limited. One of these studies, the Ontario population health survey (Wiebe et al., 1999), showed that compared to healthy individuals or those with other chronic illnesses, epilepsy patients in Ontario had more frequent visits to physicians, more hospitalizations, and greater use of psychology, and/or social work services as compared to individuals without epilepsy.

Administrative databases are easily accessible and usually comprise large subject populations (Iezzoni, 1997) providing an excellent means of studying HRU. They are felt to represent an important, untapped source of information on resource use (Langfitt, 2000). Because diagnoses are often coded in administrative databases, it is important to address the validity or accuracy of disease coding systems for the purpose of case identification. Recently, all epilepsy ICD-9 codes were validated at our health region, using emergency visits and hospitalization databases, with overall sensitivities and specificities of over 98% (Jetté et al., 2007; Reid et al., 2007).

The primary aim of this study was to provide estimates of HRU in epilepsy patients using population-based administrative databases. The secondary aim was to determine whether there were any demographic variables associated with variations in HRU. Our hypotheses were that: (1) ab-originals would be higher users of nonspecialized care and lower users of specialized care as compared to nonaboriginals; (2) urban residents would have higher visits to specialists as compared to rural residents; and (3) those of suspected lower socioeconomic status would be associated with higher HRU as compared to those of higher socioeconomic status. Knowledge about these HRU patterns is very important as it can inform funding agencies in allocative decision making for individuals with epilepsy, provide insight into possible disparities, and provide a methodological framework for further, large-scale research programs.

Methods

Databases and record linkage process

The Calgary Health Region (CHR) comprises four teaching hospitals (three adult and one children's hospital). These sites serve a population of 1.3 million people. Three Alberta administrative databases (fiscal years 1999–2001) were used for the study including: the Inpatient Discharge Abstract Database, the Alberta Physicians Claims Database, and the Emergency Abstract Database. These databases were linked using the Alberta Health Care Insurance Plan Registry (AHCIP) from the fiscal year 1999 to 2001. Based on the low health insurance fees, existent subsidy programs, and the availability of universal health care in Alberta, it has been found that the Alberta insurance registry is nearly complete and consistent, and is used as a proxy of the population of Alberta (personal communication with Dr. H. Quan). The AHCIP includes anyone registered with the provincial Alberta Health Care and creates a personal health number (PHN) that can be used to link various databases. Patient confidentiality was maintained by creating a linker variable as a substitute for the PHN. The AHCIP provides demographic information such as date of birth (DOB), gender, postal code, and markers of socioeconomic status (whether their health coverage was subsidized or not and whether or not they were on welfare or receiving assured income due to a disability). Rural postal codes were defined based on the presence of both of the following criteria: (1) if the second digit of the postal code was “0” and (2) if DMTI Spatial (a company that provides Canadian Digital Mapping Data and Geographic Information System Services) defined the postal code as a rural enumeration area centroid. Geographic Information Systems have been used increasingly for research using administrative data and have been found to provide a feasible way to study health services at the neighborhood level (Pearce et al., 2006). CHR Ethics Board approval was obtained for the study.

Study population

Epilepsy patients of all ages who had seen a physician in an outpatient setting, had visited an emergency department (ER), or had been admitted into a CHR hospital between April 1, 1999 and March 31, 2002 were identified from the various databases using the epilepsy ICD-9 code 345. Patients with an epilepsy diagnosis in the primary or any secondary positions were included. To increase the specificity of the epilepsy diagnosis, patients were only included if they had one claim (encounter) in 1999 and another claim in 2000 (see Fig. 1). The claims could be physician fee-for-service, an ER visit, or hospitalization. If they met the above criteria, then their HRU were analyzed in 2001. Only prevalent epilepsy cases were identified with this method. Researchers have suggested that to correctly identify conditions using administrative databases, at least two clinical encounters with the relevant diagnosis are needed (Soumerai et al., 1991). This is the model we, therefore, used as a guide for our case definition.

Figure 1.


Timeline for epilepsy case definition and dates used for health resource use measurements.

Patients with febrile seizures, dissociative convulsions, acquired epileptic aphasia, and convulsions not elsewhere classified were excluded. Patients who were not CHR residents or who did not receive continuous provincial health care coverage between April 1, 1999 and March 31, 2002 were also excluded.

Estimate of health care resources and other sociodemographic variables

Variables used to define HRU included physicians and ER visits, hospitalizations, length of hospital stay (LOS), and total billings for physician visits over the 1-year study period. LOS was calculated as the number of days between admission and discharge. If a patient was transferred from one CHR hospital to another during their index admission, the LOS was determined by using the admission date into the first hospital and the discharge date from the last hospital. Transfer between hospitals constituted only one hospitalization. Baseline demographic data including age and gender were also recorded. The welfare indicator in the AHCIP was used to define lower socioeconomic status. Aboriginal status was defined based on the aboriginal indicator (First Nations Flag) in the AHCIP. Any AHCIP registrant is classified as being aboriginal if they have a Band number and/or a First Nations group number. The term “status aboriginal” refers to a person who is registered under the Indian Act (Government of Canada, 1996). The Indian Act criteria have to be met for individuals to be eligible for such registration. Health Canada provides provincial health care benefits to people of aboriginal status and their dependents.

Statistical analysis

For each variable, descriptive statistics were obtained. Estimates were expressed as proportions or means with corresponding standard deviation (SD). The median was also calculated if variables were skewed. Resource utilization was calculated per patient, per year (fiscal year 2001). Age was categorized into five age groups including: (1) 0–17 years (2) 18–39 years, (3) 40–64 years, (4) 65–79 years, and (5) 80 years and above. All other variables were dichotomized including gender (female/male), location of residence (urban/rural), aboriginal status (yes/no), and welfare status (yes/no). The strength of association between independent variables and health care resource use was assessed using Fisher's exact test. Significance was set at p < 0.05. All significant variables in the univariate test were then entered in a single logistic regression model to test for associations between the independent variables and the dependent HRU variables. Odd ratios (OR), adjusted for age and gender, were calculated with 95% confidence intervals (CI).

Results

Study cohort

In the physician claims database during fiscal years 1999–2001, patients coded as having epilepsy represented 0.1% of all unique patients for which a physician claim was generated. Epilepsy patients represented 0.7% of all unique patients who had visited an ER, and 0.2% of all unique patients who had been hospitalized. Study inclusion criteria were met by 1,431 patients. Table 1 provides the descriptive characteristics of the patients. Four percent of patients were aboriginals. Aboriginals were younger than nonaboriginals with a mean age of 34.7 ± 1.95 (range 3–64) versus 37.6 ± 0.5 for nonaboriginals. There were more aboriginal males (63%) as compared to nonaboriginals (54%) and 64% of the aboriginals lived in urban areas as compared to 94% of nonaborigi- nals.

Table 1.  Baseline characteristics (n = 1,431 patients)
CharacteristicResult n = 1,431Aboriginal n = 56Nonaboriginal n = 1,375
  1. *One patient's date of birth (thus age) was missing in the database; SD, standard deviation.

Age (mean years ± SD)*Overall: 37.5 ± 17.334.7 ± 237.6 ± 0.5
Age (range in years)Overall: 2.5 − 8823.3 − 63.82.5 − 88.2
Gender n(% male)770 (53.8)6354
Aboriginal n(%)56 (3.9)
Welfare n(%)310 (21.7)n/a310 (22.5)
Subsidized n(%)227 (15.9)n/a227 (16.5)
Urban living n(%)1331 (93)6494

Health resource use

Table 2 shows that 96% of the patients had at least one outpatient office visit with a general physician (GP) or a specialist (range 0–23 visits/patient), with a mean of 2.5 visits/patient within the 1-year study period. Sixty-five percent of patients saw a GP at least once, with a mean of 1.6 visits/patient (range 0–23 visits/patient). Forty-six percent of patients saw a neurologist at least once, with a mean of 0.8 visit/patient (range 0–9 visits/patient) and 3% of patients saw a neurosurgeon (range 0–4 visits/patient). Twelve percent of patients were hospitalized (range 0–6 hospitalizations) and nearly 8% visited the ER (range 0–6 visits/patient). Total average billing costs per patient for physician visits were 112 ± 10.2 Canadian dollars per patient. This did not include the cost of emergency visits, hospitalizations, or any other direct, indirect, or intangible costs.

Table 2.  Health resource use in 1 year (n = 1,431 patients)
 naRange per patientMean ± SD per patientMediann (%) Patients with at least one visit or hospitalization
  1. SD, standard deviation. an refers to the total numbers for the particular section (e.g., total number of outpatient visits, total number of hospitalizations, total number of $) rather than referring to the total number of subjects.

Outpatient visits3,5450–23 2.5 ± 2.2 2.01,377 (96.2) 
General practitioners visits2,2140–231.6 ± 2  1.0932 (65.1)
Neurology visits1,0880–9  0.8 ± 1.10652 (45.6)
Pediatrics visits1480–8  0.1 ± 0.5071 (5) 
Neurosurgery visits 590–4 0.04 ± 0 041 (2.9)
Billings for outpatient visits ($)160,500.20 0–819112.2 ± 10.288  
Hospitalizations2580–6  0.2 ± 0.60172 (12.0)
Hospital length-of-stay (days)2,917  0–195  2 ± 120
Emergency department visits1630–6  0.1 ± 0.50111 (7.8) 

Table 3 describes HRU based on the various sociodemographic variables. Table 4 lists the OR for the utilization of health services described in Table 3. The following variables were included in our logistic regression model: age, gender, aboriginal status, and the different types of visits (outpatient visits, GP visits, neurology visits, hospitalizations, and ER visits). Younger patients (children) were more likely than older patients (adults) to see a neurologist, visit the ER, or be hospitalized (OR = 1.02, 95% C.I. 1.02–1.03, OR = 1.04, 95% C.I. 1.03–1.06, and OR = 1.01, 95% C.I. 1.01–1.02, respectively), and less likely to visit a GP (OR = 0.96, 95% C.I. 0.96–0.97). Visits to neurologists and to the ER decreased with age. Females were slightly less likely to visit a GP than males (OR = 0.8, 95% C.I. 0.63–0.99) but slightly more likely to see a neurologist (OR = 1.2, 95% C.I. 1.01–1.25).

Table 3.  Total health resource use (n = 1,431 patients) for each independent variable
 Outpatient physician visits % (n)GP visits % (n)Neurology visits % (n)ER visits % (n)Hospitalizations % (n)
  1. aOne date of birth missing in databases thus one age missing.

  2. bp < 0.01 in logistic regression analysis.

  3. cp < 0.05 in logistic regression analysis.

  4. M, male; F, female; GP, general practitioners; ER, emergency department.

Agea
 0–17 (n = 185)92 (170)18 (33)b57 (106)b21 (39)b24 (45)c
 18–39 (n = 600)98 (586)69 (415)b51 (307)b 7 (42)b10 (60)c
 40–64 (n = 552)97 (535)76 (423)b38 (208)b 5 (29)b 9 (51)c
 65–79 (n = 71)93 (66) 70 (50)b31 (22)b0 (0)b15 (11)c
 80 + (n = 22)86 (19) 46 (10)b41 (9)b  5 (1)b23 (5)c
Gender
 M (N = 770)97 (743)67 (518)c43 (333)c8 (63)11 (88) 
 F (N = 661)96 (634)63 (414)c48 (319)c7 (48)13 (84) 
Residence 
Urban (n = 1,331) 96 (1,280)65 (861) 46 (613)  8 (107)12 (161)
Rural (n = 100)97 (97) 71 (71)  39 (39)  4 (4) 11 (11) 
Aboriginal status
 Aboriginal (n = 56)89 (50)b75 (42)  23 (13)b16 (9)c27 (15)b
 Nonaboriginal (n = 1,375)  97 (1,327)b65 (890) 47 (639)b  7 (102)c 11 (157)b
 Welfare status 
 Welfare (n = 310)95 (294)68 (211) 45 (139) 9 (29)14 (44) 
 No welfare (n = 1121) 97 (1,083)64 (721) 46 (513) 7 (82)11 (128)
Table 4.  Adjusted odd ratios for the utilization of health services by age, gender, and aboriginal status
 Odds ratiosa95% CI
  1. Note:"–" = did not reach statistical significance in logistic regression model; CI = confidence interval. Age was divided into five categories as shown in Table 3.

  2. aOdds ratios mean that use of a specific resource was less likely (OR <1) or more likely (OR >1) in that particular group in relation to its comparator. For age, this table is referring to younger age (compared to older age). Thus, younger patients were less likely to see a GP (OR 0.96) but more likely to see a neurologist (OR 1.02), or be hospitalized (OR 1.01). The above odds ratio were obtained from the multivariate analysis.

Outpatient visits (overall) 
 Age
 Female gender
 Aboriginal0.3 0.12–0.74
General practitioners visits 
 Age0.960.96–0.97
 Female gender0.8 0.63–0.99
 Aboriginal
Neurology visits 
 Age1.021.02–1.03
 Female gender1.2 1.01–1.25
 Aboriginal0.3 0.17–0.61
Hospitalizations 
 Age1.011.01–1.02
 Female gender
 Aboriginal2.8 1.49–5.13
Emergency department visits 
 Age1.041.03–1.06
 Female gender
 Aboriginal2.3 1.09–4.98

Aboriginals were 3.3 times less likely to have a physician visit (any specialty; OR = 0.30, 95% C.I. 0.12–0.74) and 3.1 times less likely (OR = 0.3, 95% C.I. 0.17–0.61) to see a neurologist than nonaboriginals (Table 4). Only 23% of the aboriginals had seen a neurologist at least once in 2001 as compared to 47% of nonaboriginals (Table 3). On the other hand, aboriginals were 2.3 times more likely to visit the ER (OR = 2.3, 95% C.I. 1.09–4.98), and 2.8 times more likely to be admitted (OR = 2.8, 95% C.I. 1.49–5.13) than nonaboriginals (Table 4). Sixteen percent of the aboriginals had visited the ER, as compared to 7% of nonaboriginals, whereas 27% of the aboriginals had been admitted as compared to 11% of nonaboriginals (Table 3). Welfare status and urban residential location were not associated with higher or lower HRU. However, there seemed to be a trend toward higher GP visits and lower neurologist visits for those living in rural areas as compared to those living in urban areas. These findings may not have been significant due to a power issue (only 7% of the subjects in the study lived in rural areas).

Discussion

In this study, population-based clinical administrative databases were used to study HRU in epilepsy patients. Younger patients were more likely to see a neurologist or visit the ER, while females were slightly less likely to see a GP but more likely to see a neurologist than their male counterparts. Aboriginals were less likely to have a physician visit and see a neurologist than nonaboriginals. On the other hand, aboriginals were more likely to visit the ER or be admitted than nonaboriginals.

The hospitalization figure of 12% found in this study is in keeping with prior studies using administrative (Anonymous, 1995; Morgan & Kerr, 2004) and observational data (Thomas et al., 2001; Al-Zakwani et al., 2003, Nsengiyumva et al., 2004.) but lower than in the Ontario Health Survey (Wiebe et al., 1999). In the Ontario study, 22% of the persons with epilepsy had been admitted into a hospital in the 12 months preceding the survey. However, the survey based its figures on self-report, which can be confounded by respondent recall, and the epilepsy diagnosis was not physician-based. One advantage of administrative data is that it is based on physician diagnoses (Robinson et al., 1997). The quality of the administrative database, however, depends on the expertise of the coders and on the quality of the medical records (Kokotailo & Hill, 2005). Over 90% of coding errors relate to underreporting diagnoses and procedures rather than incorrect diagnoses (Lloyd & Rissing, 1985). An epilepsy patient may be admitted for a fracture due to a convulsive seizure with a fall but the final-coded diagnosis may only be the fracture as opposed to fracture and epilepsy. Another study also found higher rates of hospitalizations than ours as did the Ontario Health Survey, but they excluded patients with “very mild seizures” (Griffiths et al., 1999). The ER visit frequencies in this study were in keeping with prior studies (Cramer et al., 2004) but again slightly lower than in the population health survey described by Wiebe et al., (1999), likely due to the same reasons described above.

Our physician visit rates were also lower than in the population surveys by Wiebe et al., (1999) and Cramer et al., (2004). Once again, respondent recall may have biased the figures. On the other hand, many subjects with epilepsy visit their GP for reasons other than their epilepsy. In our study, less than 3% of all claims for GP visits were coded with epilepsy in the secondary diagnostic position. Only three diagnostic codes can be provided with billing claims, and most physicians only provide one diagnostic code per claim. This likely explains why the number of visits to GP was lower in our study. Another reason could be that in our study, the patient's epilepsy severity may have been milder. Only 46% of the patients saw a neurologist. However, our case definition was strict, requiring one claim in 1999 and another in 2000. Resources were then assessed in the year 2001. Thus, some of the patients could very well have seen a neurologist the year before or immediately after 2001. A longer study period would have likely shown a much higher neurologist visit rate. In fact, a prior Canadian study found that over 80% of the children with epilepsy are referred to a neurologist (Nixon Speechley et al., 1999). In addition, only prevalent cases were used. It is well known from the cost of epilepsy studies that the first year from diagnosis is most costly because of investigations, more potential for uncontrolled seizures, greater likelihood of specialist involvement, and potentially higher hospitalization rates (Morgan & Kerr, 2004).

In a cost analysis using administrative database linkage, it was found that the cost of epilepsy was greatest in the younger age group in which more than a 10-fold increase was noted in costs as compared with the nonepilepsy population (Morgan & Kerr, 2004). This is consistent with our results showing slightly more specialist visits, ER visits, and hospitalization in the younger age groups. The incidence of epilepsy follows a bimodal distribution, with more cases identified early and later in life (Hauser et al., 1991; Hauser et al., 1993). It is possible that many of the younger patients had not had epilepsy for a long time and were still being worked up or had recently started on antiepileptic drugs, requiring more health care visits than slightly older patients who may have had more well-defined longstanding epilepsy. Our administrative data could not provide us with duration and severity of epilepsy.

In this study, aboriginals were more than twice as likely than nonaboriginal to see a GP, visit the ER, or be hospitalized, yet were a third less likely to see a neurologist. Although there are no aboriginal population-based or administrative data-based epilepsy studies, there are many studies looking at HRU in aboriginals with various chronic conditions. These consistently demonstrate higher rates of GP and ER visits and hospitalizations for aboriginals as compared to nonaboriginals (Gracey & Veroni, 1995; Gruen et al., 2001; Shah et al., 2003; Cardinal et al., 2004; Gruen & Bailie, 2004; Marrie et al., 2004; Mohsin et al., 2005; Alaghehbandan et al., 2006; Kruger et al., 2006; Thomas & Anderson, 2006). One such study found very similar ratios of ER visits, GP visits, and hospitalizations as we did for First Nations individuals with diabetes, demonstrating very similar utilization rates in those with chronic conditions (Cardinal et al., 2004). Barriers to specialist care are believed to exist among aboriginals (Gracey & Veroni, 1995; Gruen & Bailie, 2004), consistent with our findings showing lower odds of seeing a neurologist than for nonaboriginals. One Canadian group studied markers of access to and quality of primary care for aboriginal people and found that although hospitalization rates were higher, procedure utilization rates were lower, suggesting insufficient or ineffective primary care (Shah et al., 2003). Possible barriers to care may include, but are not limited to, poor communication due to language differences, geographic remoteness, poverty, and cultural differences (Gruen et al., 2001). In this study, aboriginal were much more likely to live in rural areas as compared to nonaboriginals, and hence contributing to fewer neurology visits. Aboriginals may be referred to specialists but may be unable to travel to the clinics. Furthermore, many referrals to neurologists come from emergency physicians. Various studies on different conditions have shown that aboriginals walk out of emergency departments before being seen more often than nonaboriginals (Mohsin et al., 2005; Thomas & Anderson, 2006), which could, in our study, have resulted in fewer neurological referrals.

Limitations to administrative data include the inability to determine epilepsy severity, and at times to determine whether index cases are incident or prevalent (Iezzoni, 1997). It is thus likely that we underestimated the number of visits as we did not look at incident cases. Furthermore, two visits were required for patients to be included in the study, which increases specificity, but could result in missing some cases. For example, it could miss mild cases, or those whose diagnosis was very recent, not extending into the second year. It is felt that increasing the number of required visits will also lead to the identification of more severe cases, as may the selection of patients from tertiary care centers only. On the other hand, seizure frequency and severity are not the only factors in HRU. Comorbidities are a significant contributing factor in HRU. In a population-based survey, it was found that epilepsy patients with depression had higher HRU than those without depression (Cramer et al., 2004). Interestingly, they also found that the HRU was correlated with the severity of depression, but not with the severity of epilepsy (Cramer et al., 2004). It may then be equally, if not more important, to adjust for various comorbidities rather than concentrating solely on epilepsy severity.

Conclusion

In conclusion, we show that administrative data can be used successfully to study HRU in epilepsy patients, and that these analyses can uncover important variations in health care patterns within the population of patients with epilepsy. Children were found to be slightly higher health resource users, having more frequent visits to neurologists, to the emergency departments, and more frequent hospitalizations. Women were slightly more likely than men to see a neurologist but less likely to see a GP. Evidence of disparities in HRU was also demonstrated between aboriginals and nonaboriginals. Aboriginals had more frequent GP, ER, and hospitalization visits, yet significantly less specialist (neurologist) visits as compared to nonaboriginals. Although limitations exist, administrative data can provide valuable information and can be a useful screening tool to highlight areas that need to be investigated more carefully using detailed clinical information.

Acknowledgments

The authors confirm that they have read Epilepsia's position on issues involved in ethical publication and affirm that this article is consistent with those guidelines. The authors would like to thank Robyn Parker for her assistance with manuscript editing.

Conflicts of interest: The authors report no conflicts of interest.

Statistical analysis: Carried out by the first author N. Jetté in consultation with all coauthors.

Supplemental data: N/A

Ancillary