Avoidable hospitalizations in patients with systemic lupus erythematosus

Authors

  • Michael M. Ward

    Corresponding author
    1. Intramural Research Program, National Institute of Arthritis and Musculoskeletal and Skin Diseases, NIH, US Department of Health and Human Services, Bethesda, Maryland
    • NIH/NIAMS/IRP, Building 10 CRC, Room 4-1339, 10 Center Drive, MSC 1468, Bethesda, MD 20892-1468
    Search for more papers by this author

Abstract

Objective

Avoidable hospitalizations are hospitalizations for indications that could have been prevented by prompt and appropriate outpatient treatment. Avoidable hospitalizations thus serve as an indicator of poor access to, or underutilization of, medical care. This study examined risk factors for avoidable hospitalizations among patients with systemic lupus erythematosus (SLE).

Methods

Data were obtained on acute-care hospitalizations in a population-based sample of 8,670 patients with SLE age ≥18 years hospitalized in New York state in 2000, 2001, and 2002. Hospitalizations were classified as avoidable based on the principal indication for hospitalization. Patient demographic and hospital characteristics were examined as risk factors.

Results

Of 16,751 hospitalizations, 2,123 (12.7%) were for avoidable conditions, most commonly pneumonia, congestive heart failure, and cellulitis. The likelihood of avoidable hospitalizations increased progressively with age, was higher among patients with Medicare than among those with other types of medical insurance, and was higher among those of lower socioeconomic status. Hospitalizations for avoidable conditions were less likely at hospitals that admitted larger numbers of patients with SLE than at hospitals that admitted fewer patients with SLE.

Conclusion

Avoidable hospitalizations occur more commonly among older and poorer patients, suggesting that these patients have more difficulty accessing care. The lower risk of avoidable hospitalizations at centers that admit large numbers of patients with SLE may be due to patient selection or may represent better outpatient care by physicians at these centers.

INTRODUCTION

A total of 20–25% of patients with systemic lupus erythematosus (SLE) are hospitalized in a given year (1). Most hospitalizations are for the treatment of SLE or its complications, infections, or exacerbations of coexisting medical conditions (2). Hospitalizations are a major component of the total costs of care of patients with SLE (3). Many hospitalizations would be considered unavoidable because the indication for hospitalization developed acutely and was unanticipated. However, some hospitalizations may be avoidable. Avoidable hospitalizations are hospitalizations for conditions that in most cases could have been successfully treated in the outpatient setting if care had been sought in the early stages of illness or if clinical monitoring was appropriate and prompt treatment was provided (4, 5).

Avoidable hospitalizations are more common among patients with limited access to care and those without private medical insurance (4–12). Rates of avoidable hospitalizations have been accepted as indicators of inadequate access to, or inefficiency of, primary care (13). Low socioeconomic status, being an ethnic minority, older age, having many comorbid medical conditions, and living in a rural area have also been associated with increased risks of avoidable hospitalizations (5–7, 9–18). These factors may represent different patient attributes related to accessibility of care, but they may also represent groups that are more likely to have limited understanding of when or where to seek care.

Because avoidable hospitalizations represent a preventable increase in the burden of illness, carry risks of nosocomial complications, and generate unnecessary direct and indirect costs, it is important to know the frequency of avoidable hospitalizations among patients with SLE, and to identify the types of patients most commonly affected. The present study used a population-based sample of hospitalizations in New York to identify risk factors for avoidable hospitalizations in patients with SLE.

PATIENTS AND METHODS

Source of data.

Data for patients hospitalized in New York state in 2000, 2001, and 2002 were obtained from the State Department of Health Statewide Planning and Research Cooperative System. All acute-care, nonfederal hospitals are mandated to provide this agency with a discharge abstract for each hospitalization. Discharge abstracts include information on patient demographic characteristics, principal diagnosis (defined as the condition chiefly responsible for the hospitalization, by International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] codes [19]), up to 14 secondary diagnoses, hospital identifier, and attending physician identifier. The data set also included unique patient identifiers, which allowed repeat hospitalizations of individual patients to be known. However, to protect confidentiality, patients were anonymous. The abstracts were prepared from medical and billing records by trained abstractors, and were subjected to extensive reliability and consistency checks (20). Data on error rates of specific variables, such as diagnoses, were not available. In 2002, the median proportion of records with any error among all hospitals was 0.35% (20). Records with excessive error rates were returned to hospitals for correction.

For this study, all acute-care hospitalizations of patients age ≥18 years for whom the principal diagnosis or any of the 14 secondary diagnoses was SLE were identified. Of the 23,197 hospitalizations, we excluded 6,031 hospitalizations that were elective admissions (n = 4,730), were for childbirth (n = 519), resulted from an interhospital transfer (n = 573), or were missing the patient identifier (n = 209; 5 cases had more than 1 reason for exclusion). Also excluded were 198 hospitalizations of patients who resided outside of New York, because these patients may have been away from their usual source of care, and 217 hospitalizations for which information on the patient's residence was missing. The remaining set included 16,751 emergency or urgent hospitalizations in 8,670 patients.

The study protocol was exempted from human subjects review by the National Institutes of Health Office of Human Subjects Protection.

Study variables.

The definition of an avoidable hospitalization was based on whether 1 of 12 specific diagnoses was listed as the primary indication for hospitalization, as developed by Weissman and colleagues (4). Avoidable hospitalizations were identified as those for which the principal discharge diagnosis (by ICD-9-CM code) was ruptured appendix, asthma, cellulitis, congestive heart failure, diabetes mellitus out of control (ketoacidosis, hyperosmolar coma, or hypoglycemic coma), gangrene, hypokalemia, common communicable diseases for which immunizations exist, malignant hypertension, pneumonia, pyelonephritis, and perforated or bleeding ulcer. These diagnoses were chosen by a physician panel as largely preventable complications that were clinically unambiguous and that had specific diagnostic codes. This definition has been used in previous studies (7, 9, 10, 17). The principal discharge diagnosis is used rather than the admitting diagnosis because the correct diagnosis may only become evident after in-hospital evaluation (21, 22).

The patient-related independent variables were age, sex, race, Hispanic ethnicity, type of medical insurance (private insurance, Medicare, public insurance other than Medicare, no insurance, or unknown type), residence in a rural county, and socioeconomic status. Because the data set did not include measures of socioeconomic status, patients were assigned a socioeconomic status score based on the socioeconomic characteristics of their zip code of residence, using a previously validated approach (23, 24). First, using principal components analysis of socioeconomic indicators from the 2000 US Census files (http://factfinder.census.gov), 7 measures were identified to be included in a composite measure of socioeconomic status (log of median household income, proportion with income <200% of the federal poverty level, log of median house value, log of median monthly rent, mean education level, proportion of people age ≥25 years who were college graduates, and proportion of employed persons with a professional occupation). Each of these measures loaded strongly on a single factor, with all factor loads >0.75, and together explained 70% of the variance across all US zip codes. Second, means and SDs were computed for each measure among all zip codes in New York, along with corresponding Z scores for each zip code. The socioeconomic status score was then computed as the sum of the Z scores for all 7 measures. Comorbid conditions were not included as potential risk factors because many of the avoidable hospitalizations represent complications of comorbid conditions, and including these conditions as risk factors would therefore be circular.

Because the type of hospital to which patients were admitted may have been associated with whether the hospitalization was avoidable or not, several hospital-related variables were also examined. These variables were the number of beds, teaching status (whether or not the hospital sponsored an approved residency program), whether or not the hospital was an academic medical center, and hospital and physician volume, based on the annual number of patients with SLE admitted. Hospitals were classified as high volume if they averaged >50 emergency or urgent hospitalizations of patients with SLE per year (25). Physicians were categorized into 3 groups (<1 hospitalization of a patient with SLE per year, 1–3 hospitalizations per year, or >3 hospitalizations per year) based on their average annual number of urgent or emergency hospitalizations of patients with SLE over the 3-year period (26).

Statistical analysis.

The unit of analysis was the hospitalization. Logistic regression analysis was used to determine the association between each independent variable and avoidable hospitalization. Age and socioeconomic status score were transformed to categorical variables to allow for nonlinear associations. Multivariable analyses were performed that tested all patient-related and hospital-related variables. Reduced models were then tested using only those independent variables that were significantly associated with the risk of avoidable hospitalization in the full models. The analyses were performed using generalized estimating equations to account for repeated hospitalizations of individual patients (27). Goodness-of-fit was tested for all multivariable models, and these tests did not indicate lack of fit (all P > 0.68) (28). Analyses were performed using SAS software, version 9.1 (SAS Institute, Cary, NC). All hypothesis tests were 2-tailed, and P values less than 0.05 were considered statistically significant.

RESULTS

The study included 16,751 hospitalizations in 8,670 patients. Eighty-eight percent of the patients were women, 52% were white, 29% were black, 1.7% were Asian/Pacific Islander, 0.2% were Native American, 12% reported other race, and race was unknown for 5%. Ten percent reported Hispanic ethnicity, and information on Hispanic ethnicity was missing for 14.6% of patients. Five percent lived in rural counties. Patients were hospitalized at 229 hospitals, including 68 teaching hospitals, 19 academic medical centers, and 37 high-volume hospitals. Twenty-seven percent of teaching hospitals were academic medical centers. Eighty-one percent of high-volume hospitals were teaching hospitals, and 49% of high-volume hospitals were academic medical centers.

Of the 16,751 hospitalizations, 2,123 (12.7%) were classified as avoidable (Table 1). Avoidable hospitalizations occurred in 1,666 patients (19.2%). Of the 5,326 patients with only 1 hospitalization, the hospitalization was avoidable in 11.9%.

Table 1. Frequency of avoidable hospitalizations by diagnosis
DiagnosisAvoidable hospitalizations, % (n = 2,123)Total hospitalizations, % (n = 16,751)
Ruptured appendix0.80.1
Asthma2.20.3
Cellulitis19.32.4
Congestive heart failure23.43.0
Diabetes mellitus out of control2.20.3
Gangrene0.30.04
Hypokalemia0.60.08
Communicable diseases0.00.0
Malignant hypertension2.90.4
Pneumonia40.15.1
Pyelonephritis5.70.7
Perforated or bleeding ulcer2.50.3

In univariable analyses, avoidable hospitalizations occurred more frequently among older patients, those with Medicare, and those with lower socioeconomic status scores (Table 2). In addition, avoidable hospitalizations were more frequent among patients hospitalized at smaller hospitals and at hospitals that were not teaching hospitals, hospitals that were not academic medical centers, and at low-volume hospitals.

Table 2. Univariable associations between patient and hospital characteristics and risk of avoidable hospitalization, by logistic regression analysis*
Independent variableAvoidable hospitalizations, %OR95% CIP
  • *

    The ORs present the relative odds that patients with the characteristic would have a hospitalization for an avoidable condition compared with patients without the characteristic. OR = odds ratio; 95% CI = 95% confidence interval; SES = socioeconomic status.

Patient characteristics    
 Age <35 years9.31.00 (reference)
 Age 35–44 years11.31.231.03–1.470.02
 Age 45–54 years12.51.391.17–1.660.0001
 Age 55–64 years13.71.551.29–1.87< 0.0001
 Age ≥65 years17.22.021.71–2.39< 0.0001
 Men13.51.00 (reference)
 Women12.60.920.78–1.090.34
 White13.01.00 (reference)
 Black12.60.960.85–1.090.56
 Asian/Pacific Islander10.90.830.55–1.230.35
 Native American11.80.900.30–2.660.85
 Other race12.90.990.83–1.180.95
 Unknown race10.60.800.62–1.020.07
 Not Hispanic12.91.00 (reference)
 Hispanic11.50.870.72–1.050.15
 Hispanic ethnicity unknown12.40.950.81–1.110.52
 Private medical insurance10.61.00 (reference)
 Medicare15.21.511.33–1.71< 0.0001
 Public insurance11.71.120.97–1.310.12
 No insurance11.81.130.85–1.490.39
 Unknown insurance9.10.840.35–2.020.71
 Highest quartile of SES score12.21.00 (reference)
 Third quartile of SES score12.41.020.87–1.190.83
 Second quartile of SES score12.31.010.86–1.180.88
 Lowest quartile of SES score13.81.150.99–1.340.06
 Urban residence12.61.00 (reference)
 Rural residence14.31.160.93–1.450.18
Hospital characteristics    
 Beds ≥50011.21.00 (reference)
 Beds 300–49913.31.221.08–1.390.002
 Beds 100–29914.11.301.14–1.48< 0.0001
 Beds <10015.11.411.06–1.870.02
 Nonteaching hospital13.61.00 (reference)
 Teaching hospital12.10.870.78–0.970.008
 Community hospital13.31.00 (reference)
 Academic medical center11.30.830.73–0.940.002
 Low-volume hospital14.11.00 (reference)
 High-volume hospital11.30.770.69–0.86< 0.0001
 High-volume physician12.21.00 (reference)
 Moderate-volume physician13.51.120.96–1.310.13
 Low-volume physician12.10.990.85–1.160.93

In a multivariable analysis, the risk of avoidable hospitalization increased progressively with age, was higher among patients with Medicare as the type of insurance, and was higher among those in the lowest quartile of socioeconomic status (Table 3). There was no difference in risk among patients in the 3 highest socioeconomic status quartiles. In addition, patients who reported their race as other than white, black, Asian/Pacific Islander, or Native American had a somewhat higher risk of avoidable hospitalizations than whites. Hospital size was not associated with the risk of avoidable hospitalization, but being hospitalized at a high-volume hospital was associated with a 13% lower risk. Teaching status and whether or not the hospital was an academic medical center were not included because hospitals in these categories overlapped with those classified according to SLE-related hospital volume, and SLE-related hospital volume was more strongly associated with the frequency of avoidable hospitalizations in the univariable analysis (χ2 = 24.5 for SLE-related hospital volume versus 7.1 for teaching status and 10.1 for academic medical center). Associations were similar in the reduced model that excluded variables that were not significantly associated with the risk of avoidable hospitalizations, except that race was no longer associated with the risk of avoidable hospitalizations (Table 3). The association between Medicare and risk of avoidable hospitalizations was also present in an analysis restricted to patients age <65 years (adjusted odds ratio 1.27, 95% confidence interval 1.09–1.49, P = 0.003) and was not modified by adjustment for the presence of chronic renal failure.

Table 3. Multivariable associations between patient and hospital characteristics and the risk of avoidable hospitalizations, by logistic regression analysis*
Independent variableModel including patient and hospital-related variablesReduced model
OR95% CIPOR95% CIP
  • *

    See Table 2 for definitions.

  • SES score dichotomized for this model into lowest quartile versus top 3 quartiles.

Age <35 years1.001.00
Age 35–44 years1.211.01–1.460.041.221.01–1.460.04
Age 45–54 years1.381.15–1.650.00031.381.16–1.650.0003
Age 55–64 years1.561.29–1.89< 0.00011.571.29–1.90< 0.0001
Age ≥65 years1.821.49–2.23< 0.00011.841.50–2.25< 0.0001
Men1.00   
Women0.940.79–1.110.44   
White1.001.00
Black1.100.96–1.270.141.110.98–1.270.09
Asian/Pacific Islander0.980.65–1.480.940.990.66–1.490.99
Native American1.490.57–3.910.421.490.58–3.830.41
Other race1.241.01–1.530.041.150.96–1.380.13
Unknown race0.930.70–1.240.630.890.69–1.150.38
Not Hispanic1.00   
Hispanic0.840.68–1.050.13   
Hispanic ethnicity unknown0.950.79–1.140.57   
Private medical insurance1.001.00
Medicare1.271.10–1.470.0011.271.10–1.470.001
Public insurance1.150.98–1.340.091.130.97–1.330.11
No insurance1.140.86–1.510.361.140.86–1.510.37
Unknown insurance0.800.36–1.790.600.800.35–1.830.61
Highest quartile of SES score1.001.00
Third quartile of SES score1.020.87–1.190.811.00
Second quartile of SES score1.030.87–1.210.731.00
Lowest quartile of SES score1.181.01–1.390.031.161.02–1.310.02
Urban residence1.00   
Rural residence1.010.79–1.300.93   
Beds ≥5001.00   
Beds 300–4991.110.95–1.290.17   
Beds 100–2991.090.91–1.310.35   
Beds <1001.070.75–1.530.69   
Low-volume hospital1.001.00
High-volume hospital0.870.75–1.000.050.820.73–0.910.0002

Because infections may occur precipitously in patients treated with immunosuppressive medications and therefore may not be avoidable, the analyses were repeated after excluding the 1,382 hospitalizations due to pneumonia, cellulitis, or pyelonephritis (Table 4). This analysis included 741 avoidable hospitalizations in 593 patients. Older age, having Medicare as the type of insurance, and being in the lowest socioeconomic status score quartile were also associated with a higher risk of noninfection-related avoidable hospitalization, and being hospitalized at a high-volume hospital was associated with a 23% lower risk. Using this definition of avoidable hospitalizations, blacks were at an increased risk compared with whites, and Hispanics were at a lower risk compared with non-Hispanics.

Table 4. Multivariable associations between patient and hospital characteristics and the risk of avoidable hospitalizations, by logistic regression analysis, excluding hospitalizations for pneumonia, cellulitis, and pyelonephritis*
Independent variableModel including patient and hospital-related variablesReduced model
OR95% CIPOR95% CIP
  • *

    NE = not estimable due to small numbers; see Table 2 for additional definitions.

  • SES score dichotomized for this model into lowest quartile versus top 3 quartiles.

Age <35 years1.001.00
Age 35–44 years1.120.77–1.620.541.120.77–1.620.56
Age 45–54 years1.801.26–2.570.0021.801.26–2.580.002
Age 55–64 years2.531.75–3.65< 0.00012.531.75–3.66< 0.0001
Age ≥65 years3.332.25–4.93< 0.00013.312.24–4.90< 0.0001
Men1.00   
Women0.900.67–1.200.47   
White1.001.00
Black1.281.03–1.600.031.291.04–1.610.02
Asian/Pacific Islander1.260.65–2.420.491.260.66–2.400.48
Native AmericanNE  NE  
Other race1.390.95–2.030.091.390.96–2.020.08
Unknown race1.130.66–1.910.661.110.66–1.890.69
Not Hispanic1.001.00
Hispanic0.610.40–0.930.020.620.41–0.950.03
Hispanic ethnicity unknown0.780.55–1.110.170.800.56–1.130.20
Private medical insurance1.001.00
Medicare1.361.06–1.750.021.361.06–1.750.02
Public insurance1.110.83–1.490.481.110.83–1.490.48
No insurance1.050.61–1.800.851.050.62–1.790.86
Unknown insurance1.200.35–4.070.771.210.34–4.340.77
Highest quartile of SES score1.001.00
Third quartile of SES score1.020.78–1.330.871.00
Second quartile of SES score1.010.77–1.310.951.00
Lowest quartile of SES score1.301.01–1.680.041.271.04–1.540.02
Urban residence1.00   
Rural residence1.090.73–1.610.68   
Beds ≥5001.00   
Beds 300–4991.100.86–1.410.44   
Beds 100–2991.020.75–1.370.91   
Beds <1000.670.35–1.260.22   
Low-volume hospital1.001.00
High-volume hospital0.770.61–0.980.030.770.63–0.920.005

DISCUSSION

In this population-based study, avoidable hospitalizations accounted for 12.7% of all urgent or emergency hospitalizations of patients with SLE. This proportion is comparable with those reported in surveys of hospitalizations in adults without SLE, which ranged from 8% to 15% (9, 10, 12). Infections and congestive heart failure comprised the majority of avoidable hospitalizations. Excluding infections, avoidable hospitalizations accounted for 4.8% of urgent or emergency hospitalizations of patients with SLE.

Avoidable hospitalizations affected nearly 20% of patients, and were not concentrated among a small minority of patients who were repeatedly hospitalized for the same condition. Despite this, some patient-related factors were identified as being associated with an increased risk of avoidable hospitalization. Consistent with prior research (5, 6, 9, 10, 12, 14–17), avoidable hospitalizations were more likely to occur among older patients and those of lower socioeconomic status; blacks were at a higher risk for noninfectious avoidable hospitalizations. The risk of avoidable hospitalizations increased progressively with age, possibly because these patients had more comorbid conditions, greater difficulty managing more complicated medication regimens or obtaining medications, or less social support. Living in a poorer neighborhood was also a risk factor for avoidable hospitalizations, although the increased risk was limited to patients in the lowest quartile of socioeconomic status score. Low socioeconomic status may adversely affect access to care or access to needed medications. However, low socioeconomic status has been associated with an increased risk of avoidable hospitalizations even in countries with universal financial access to care, suggesting that organizational barriers to care, limited patient education, or differences in care-seeking behavior may contribute to avoidable hospitalizations among persons of lower socioeconomic status (29).

In contrast to other studies (4–6, 9, 11, 12), the risk of avoidable hospitalization was not increased among patients without medical insurance or those with Medicaid. It is possible that having a chronic illness such as SLE caused patients to seek out and establish resources for care that mitigated the risk of avoidable hospitalization associated with being uninsured or having Medicaid in other studies. However, having Medicare was a risk factor, over and above any association with age or end-stage renal disease. Medicare may be obtained by patients receiving Social Security Disability Insurance, and thus, in this group, Medicare may serve as an indicator of severity of illness. There was no increase in risk among patients living in rural areas, possibly because rural areas of New York are not as geographically isolated as in other regions.

Previous studies of risk factors for avoidable hospitalizations have not examined associations with the type of hospital to which patients were admitted. In this study, hospitalizations at high-volume hospitals, academic medical centers, and teaching hospitals were less likely to be for avoidable diagnoses than were hospitalizations at low-volume hospitals, community hospitals, or nonteaching hospitals. Adjusting for differences in patient characteristics, hospitalizations at centers that admitted larger numbers of patients with SLE were 18–23% less likely to be for an avoidable diagnosis than were hospitalizations at centers that admitted fewer patients with SLE. There are several possible explanations for this association. First, this association may be a manifestation of differences in the roles of primary care and secondary care hospitals on one hand and tertiary care hospitals on the other hand. Primary care and secondary care hospitals might be expected to treat a larger proportion of patients with common acute problems. Second, patients treated at high-volume centers (or academic centers or teaching hospitals) may differ from other patients in their health behavior. Electing to be treated at a high-volume hospital may be a marker of patients who are more actively involved in their care or who have a greater propensity to seek care, which might influence the likelihood of an avoidable hospitalization. Third, high-volume hospitals (or academic centers or teaching hospitals) may have physicians who provide higher-quality outpatient care. The pool of patients admitted to these hospitals may therefore be less likely to require hospitalization for a condition that was preventable by good outpatient management. All 3 explanations may have some role. Unfortunately, we cannot determine from the available data what role these factors may have had in mediating the association between hospital volume and risk of avoidable hospitalizations. We also cannot exclude the possibility that discharge diagnoses in patients with SLE were coded differently in low-volume and high-volume hospitals.

It is important to draw a distinction between avoidable hospitalizations and inappropriate hospitalizations. Inappropriate hospitalizations are those in which the severity or acuteness of the problem does not require inpatient care or for which other health care settings might be more effective. Avoidable hospitalizations are those for which the circumstances leading to the hospitalization, if altered, might have eliminated the need for hospitalization. Hospitalizations may be appropriate for a given presentation, but might still have been avoidable.

The strengths of the study include a large population-based sample, assessment of both patient-related and hospital-related risk factors, and examination of the influence of infection-related hospitalizations on risk factors. The study is limited in that our results were based on discharge diagnosis codes, which may not have been accurate or complete for all patients. We could not confirm that all patients met current classification criteria for SLE. In administrative databases, undercoding is more common than overcoding, although studies of inpatient databases have reported sensitivities of >0.88 and specificities approaching 1.00 for diagnoses of heart, lung, liver, and renal diseases (30–32). We did not have patient-level data on socioeconomic status, but the area-based measure that was used was validated against patient-level data (24). The study is also limited in using avoidable hospitalizations to infer problems in access to outpatient care, rather than directly examining problems with access. This limitation is shared with all other studies that use this outcome, and is offset somewhat by the representative samples included in these studies. In addition, the study does not permit causal associations to be made in relating risk factors to avoidable hospitalizations.

Avoidable hospitalizations among patients with SLE are more likely to occur among older and poorer patients, and those with Medicare as their type of medical insurance. This finding may indicate poorer access to primary care for these patients, poorer monitoring of ongoing conditions, or delays in seeking care. To the extent that rheumatologists serve as primary care providers for their patients with SLE, this may reflect suboptimal access to care from rheumatologists. Identifying the processes leading to avoidable hospitalizations in these patients may suggest strategies to reduce preventable illness.

AUTHOR CONTRIBUTIONS

Dr. Ward 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.

Study design. Ward.

Acquisition of data. Ward.

Analysis and interpretation of data. Ward.

Manuscript preparation. Ward.

Statistical analysis. Ward.

Ancillary