Background

Fifteen percent of all emergency department (ED) visits and 40% of hospital admissions through the ED in the United States are for patients aged 65 years or older. Reasons that older adults visit the ED vary from experiencing a health problem that demands immediate attention to having difficulty accessing ambulatory care. Although addressing a health need is the most common and pressing reason older adults visit the ED, factors related to access to or their experience of care may also be related to their use of the ED. For example, the presence of a usual care physician has been shown to decrease an older adult’s risk of visiting the ED.

Three main treatment patterns are possible when older adults present at an ED—they can be released, kept for an observation stay of shorter than 48 hours, or admitted to the hospital for further treatment. Decisions around these treatment patterns have been studied in acute cardiac disease but less so in other conditions. Severity of the presenting condition certainly contributes to which of these treatment patterns a patient experiences, but other factors may also be important. If a patient is being closely managed by an ambulatory care physician, he or she may present at the ED only for a serious issue necessitating hospitalization. Likewise, an observation stay or admission may not be necessary if a patient can clearly identify a physician who manages his or her care in the ambulatory setting. As a result, continuity of ambulatory care may influence whether a patient presents at the ED and the ensuing treatment pattern of the ED episode.

Importance

The fee-for-service Medicare program provides health care coverage for most older adults in the United States. Fee-for-service Medicare patients can visit any physician willing to treat them, and physicians can refer patients without network restrictions, with the typical Medicare patient experiencing 8 visits with 4 physicians annually. The pattern of visits provides insight into a patient’s care, with a visit pattern concentrated around a physician or small number of physicians suggesting more continuity in contrast to a pattern diffused across several physicians, which suggests greater fragmentation. Evidence from the visit patterns of older adults in Canada, as well as commercially insured children and Medicaid patients in the United States, suggests that a high level of continuity is related to less ED utilization, but older adults in fee-for-service Medicare have not been studied.

Goals of This Investigation

We expected that older adults with higher continuity of ambulatory visits would have a lower rate of ED utilization. We analyzed claims data from a large, national sample of fee-for-service Medicare beneficiaries to study whether the continuity of visits for older adults, using 2 different metrics of continuity, had any relationship with the occurrence of an ED episode. We then examined whether continuity was related to whether a patient has an observation stay or hospital admission compared with being discharged home from the ED.

Study Design

This study used a survival analysis of Medicare patients’ ambulatory visit patterns, starting in 2011 for up to 24 months. Patients were followed for 24 months or until death or occurrence of the composite endpoint of an ED episode; that is, an ED visit followed by discharge, an observation stay, or a hospital admission through the ED. Because secondary, deidentified data were used, this study was exempt from institutional review board approval in accordance with Federal common rule (section 45 CFR 46.101[b][5]).

Selection of Participants

Continuously enrolled fee-for-service Medicare beneficiaries aged 66 years or older by the end of 2011 from a 20% random sample were identified in the Master Beneficiary Summary File in the Chronic Conditions Data Warehouse. Medicare Advantage beneficiaries were excluded because their claims data are not available to researchers. Beneficiaries with fewer than 4 ambulatory evaluation and management visits in 2011 were also excluded because continuity metrics can too easily reach their minimum or maximum values of 0 or 1 with few visits, which would create bias in the results. Approximately 30% of the otherwise eligible population were excluded on this basis (characteristics shown in Table E1 , available online at http://www.annemergmed.com ).

Methods of Measurement

Ambulatory visits in Medicare are billed in the Carrier file or, for some providers who work at Federally Qualified Health Centers or Rural Health Clinics, in the hospital outpatient file. In the Carrier file, ambulatory evaluation and management codes billed by primary care and specialist physicians, as identified by the specialty code on the claim, were included in the measurement of continuity (evaluation and management codes in Appendix E1 , available online at http://www.annemergmed.com ). In the hospital outpatient file, visits at Federally Qualified Health Centers or Rural Health Clinics were identified by the facility type 7 and service classification type of either 3 or 1. For these visits, which constituted less than 2% of the total visits used to measure continuity, there is no unique identifier for provider specialty, so visits to each unique center were effectively considered visits to the same physician. In both files, only 1 visit per day per physician for each patient was counted in the measurement of continuity of care.

Two continuity metrics were used, the Continuity of Care (COC) score and the Usual Provider Continuity (UPC) score. The COC score was the primary continuity metric because it uses total number of visits, total number of physicians, and number of visits with each physician to measure the dispersion of a patient’s total visit pattern. The UPC score was used as a secondary metric because it captures the concentration of visits to a single physician but does not account for the dispersion of visits across all physicians in a patient’s visit pattern (illustration in Appendix E1 , available online at http://www.annemergmed.com ).

Outcome Measures

The main outcome was the occurrence of an ED episode, which was a composite outcome defined as an ED visit and discharge, observation stay through the ED, or hospital admission through the ED. Each of the 3 types of ED episodes was identified separately in claims data. ED visits that result in discharge (treat and release) are recorded in outpatient revenue center claims, and hospitalizations through the ED are in inpatient revenue center claims (codes 0450 to 0459 or 0981); observation stays are recorded in either outpatient or inpatient revenue center claims (code 0762). Any ED episode with a trauma International Classification of Diseases, Ninth Revision diagnosis code (959) was excluded because occurrence of trauma is unlikely related to a patient’s ambulatory care.

Analysis

After a patient had 4 visits in 2011, he or she was observed for up to 24 months. Each continuity score was measured until the patient presented at an ED, died, or reached the end of his or her 24-month observation period in 2013, whichever occurred first. The COC and UPC scores are measured on a scale from 0 (lowest continuity) to 1 (highest continuity); their values were multiplied by 10 in the statistical models so that the regression results could be interpreted relative to 0.1-unit, or 10%, intervals.

Time-dependent Cox proportional hazards regression was used to perform the survival analysis. We used survival analysis because it censors a patient from further observation once an ED episode occurs and avoids the confounding that could occur with measuring continuity throughout the time a patient could experience multiple ED episodes. As exposure time increases, continuity changes as visits accumulate, which necessitated treating continuity as a time-dependent variable. The continuity scores were cumulatively measured monthly from the fourth visit while staying the same if a patient had no visits in a given month. The time-dependent Cox model reads each observation distinctly each month, thereby avoiding temporal autocorrelation. Separate models were run for COC and UPC measures, with the outcome being the composite ED episode. Less than 1% of the beneficiaries had variables with missing values and were dropped from the multivariate analyses.

As a secondary analysis among beneficiaries with an ED episode, we tested whether continuity was associated with the type of ED episode to provide insights into how continuity of care might be related to different treatment patterns once a patient presents at the ED. The COC and UPC were measured during the 12 months preceding the ED episode. Multinomial logistic regression was used to estimate the relative risk of an observation stay or hospitalization through the ED relative to an ED visit alone.

Several additional variables were used in the analyses to control for factors that may be related to continuity of care or ED utilization. Demographic and enrollment characteristics were gleaned from the Master Beneficiary Summary File. Beneficiary age was a discrete variable and represented the beneficiary’s age at the end of 2011. Sex was coded as female versus male. Race and ethnicity was coded as non-Hispanic white, black, Hispanic, Asian, or other. Beneficiaries whose Medicare coverage preceded turning aged 65 years were considered disabled. Those who were Medicaid dual eligible any month between 2011 and 2013 were deemed dual-eligible beneficiaries.

Patient illness burden at baseline was accounted for in more than 1 way. First, each patient’s 2011 Hierarchical Condition Category (HCC) score was included in the models because it is a numeric value specifically developed for Medicare patients to capture expected resource use according to a patient’s clinical profile in the previous year (2010). HCC scores were divided into quartiles according to their data distribution. In addition, because sicker patients tend to need more medical care, the total number of ambulatory evaluation and management visits and total number of ED episodes in the 12 months before the start of a patient’s study observation period were included as 2 distinct variables in the Cox models.

For the secondary analysis, the same covariates from the main analysis were included in the model, although some modifications were necessary to control for illness burden because this analysis was not time dependent. First, total ambulatory evaluation and management visits were measured during the 12 months before the ED episode. Second, because hospitalization in the previous 30 days is a risk factor for ED utilization, an indicator variable marked whether a hospitalization occurred in the month before the ED episode. Finally, another indicator variable was added for whether the ED episode began on the weekend (Saturday or Sunday), which may influence whether a patient is discharged home from the ED, observed through the ED, or admitted through the ED.

Finally, hospital referral region fixed effects were used to account for any time-invariant geographic factors such as ED availability or hospital bed capacity that could affect the relationship between continuity and ED utilization.

In sensitivity analyses, the COC score and UPC score were each run with mortality as an outcome in bivariate Cox proportional hazards regression models to determine whether the censoring of deceased patients might confound the main results by leaving healthier patients in the analysis. Models were also run with lagged COC or UPC score values 1, 3, 6, or 12 months, rather than the value for the immediately preceding month. For a 1-month lag, analyses were restricted to beneficiaries with study observation periods of up to 23 months, with a 3-month lag requiring up to 21 months, a 6-month lag requiring up to 18 months, and a 12-month lag requiring up to 12 months.

Analyses were conducted with SAS-EG (version 7.1; SAS Institute, Inc., Cary, NC), with the Efron option used in the Cox proportional hazards regression models to adjust for tied events.

Study Design

This study used a survival analysis of Medicare patients’ ambulatory visit patterns, starting in 2011 for up to 24 months. Patients were followed for 24 months or until death or occurrence of the composite endpoint of an ED episode; that is, an ED visit followed by discharge, an observation stay, or a hospital admission through the ED. Because secondary, deidentified data were used, this study was exempt from institutional review board approval in accordance with Federal common rule (section 45 CFR 46.101[b][5]).

Selection of Participants

Continuously enrolled fee-for-service Medicare beneficiaries aged 66 years or older by the end of 2011 from a 20% random sample were identified in the Master Beneficiary Summary File in the Chronic Conditions Data Warehouse. Medicare Advantage beneficiaries were excluded because their claims data are not available to researchers. Beneficiaries with fewer than 4 ambulatory evaluation and management visits in 2011 were also excluded because continuity metrics can too easily reach their minimum or maximum values of 0 or 1 with few visits, which would create bias in the results. Approximately 30% of the otherwise eligible population were excluded on this basis (characteristics shown in Table E1 , available online at http://www.annemergmed.com ).

Methods of Measurement

Ambulatory visits in Medicare are billed in the Carrier file or, for some providers who work at Federally Qualified Health Centers or Rural Health Clinics, in the hospital outpatient file. In the Carrier file, ambulatory evaluation and management codes billed by primary care and specialist physicians, as identified by the specialty code on the claim, were included in the measurement of continuity (evaluation and management codes in Appendix E1 , available online at http://www.annemergmed.com ). In the hospital outpatient file, visits at Federally Qualified Health Centers or Rural Health Clinics were identified by the facility type 7 and service classification type of either 3 or 1. For these visits, which constituted less than 2% of the total visits used to measure continuity, there is no unique identifier for provider specialty, so visits to each unique center were effectively considered visits to the same physician. In both files, only 1 visit per day per physician for each patient was counted in the measurement of continuity of care.

Two continuity metrics were used, the Continuity of Care (COC) score and the Usual Provider Continuity (UPC) score. The COC score was the primary continuity metric because it uses total number of visits, total number of physicians, and number of visits with each physician to measure the dispersion of a patient’s total visit pattern. The UPC score was used as a secondary metric because it captures the concentration of visits to a single physician but does not account for the dispersion of visits across all physicians in a patient’s visit pattern (illustration in Appendix E1 , available online at http://www.annemergmed.com ).

Outcome Measures

The main outcome was the occurrence of an ED episode, which was a composite outcome defined as an ED visit and discharge, observation stay through the ED, or hospital admission through the ED. Each of the 3 types of ED episodes was identified separately in claims data. ED visits that result in discharge (treat and release) are recorded in outpatient revenue center claims, and hospitalizations through the ED are in inpatient revenue center claims (codes 0450 to 0459 or 0981); observation stays are recorded in either outpatient or inpatient revenue center claims (code 0762). Any ED episode with a trauma International Classification of Diseases, Ninth Revision diagnosis code (959) was excluded because occurrence of trauma is unlikely related to a patient’s ambulatory care.

Analysis

After a patient had 4 visits in 2011, he or she was observed for up to 24 months. Each continuity score was measured until the patient presented at an ED, died, or reached the end of his or her 24-month observation period in 2013, whichever occurred first. The COC and UPC scores are measured on a scale from 0 (lowest continuity) to 1 (highest continuity); their values were multiplied by 10 in the statistical models so that the regression results could be interpreted relative to 0.1-unit, or 10%, intervals.

Time-dependent Cox proportional hazards regression was used to perform the survival analysis. We used survival analysis because it censors a patient from further observation once an ED episode occurs and avoids the confounding that could occur with measuring continuity throughout the time a patient could experience multiple ED episodes. As exposure time increases, continuity changes as visits accumulate, which necessitated treating continuity as a time-dependent variable. The continuity scores were cumulatively measured monthly from the fourth visit while staying the same if a patient had no visits in a given month. The time-dependent Cox model reads each observation distinctly each month, thereby avoiding temporal autocorrelation. Separate models were run for COC and UPC measures, with the outcome being the composite ED episode. Less than 1% of the beneficiaries had variables with missing values and were dropped from the multivariate analyses.

As a secondary analysis among beneficiaries with an ED episode, we tested whether continuity was associated with the type of ED episode to provide insights into how continuity of care might be related to different treatment patterns once a patient presents at the ED. The COC and UPC were measured during the 12 months preceding the ED episode. Multinomial logistic regression was used to estimate the relative risk of an observation stay or hospitalization through the ED relative to an ED visit alone.

Several additional variables were used in the analyses to control for factors that may be related to continuity of care or ED utilization. Demographic and enrollment characteristics were gleaned from the Master Beneficiary Summary File. Beneficiary age was a discrete variable and represented the beneficiary’s age at the end of 2011. Sex was coded as female versus male. Race and ethnicity was coded as non-Hispanic white, black, Hispanic, Asian, or other. Beneficiaries whose Medicare coverage preceded turning aged 65 years were considered disabled. Those who were Medicaid dual eligible any month between 2011 and 2013 were deemed dual-eligible beneficiaries.

Patient illness burden at baseline was accounted for in more than 1 way. First, each patient’s 2011 Hierarchical Condition Category (HCC) score was included in the models because it is a numeric value specifically developed for Medicare patients to capture expected resource use according to a patient’s clinical profile in the previous year (2010). HCC scores were divided into quartiles according to their data distribution. In addition, because sicker patients tend to need more medical care, the total number of ambulatory evaluation and management visits and total number of ED episodes in the 12 months before the start of a patient’s study observation period were included as 2 distinct variables in the Cox models.

For the secondary analysis, the same covariates from the main analysis were included in the model, although some modifications were necessary to control for illness burden because this analysis was not time dependent. First, total ambulatory evaluation and management visits were measured during the 12 months before the ED episode. Second, because hospitalization in the previous 30 days is a risk factor for ED utilization, an indicator variable marked whether a hospitalization occurred in the month before the ED episode. Finally, another indicator variable was added for whether the ED episode began on the weekend (Saturday or Sunday), which may influence whether a patient is discharged home from the ED, observed through the ED, or admitted through the ED.

Finally, hospital referral region fixed effects were used to account for any time-invariant geographic factors such as ED availability or hospital bed capacity that could affect the relationship between continuity and ED utilization.

In sensitivity analyses, the COC score and UPC score were each run with mortality as an outcome in bivariate Cox proportional hazards regression models to determine whether the censoring of deceased patients might confound the main results by leaving healthier patients in the analysis. Models were also run with lagged COC or UPC score values 1, 3, 6, or 12 months, rather than the value for the immediately preceding month. For a 1-month lag, analyses were restricted to beneficiaries with study observation periods of up to 23 months, with a 3-month lag requiring up to 21 months, a 6-month lag requiring up to 18 months, and a 12-month lag requiring up to 12 months.

Analyses were conducted with SAS-EG (version 7.1; SAS Institute, Inc., Cary, NC), with the Efron option used in the Cox proportional hazards regression models to adjust for tied events.

Characteristics of Study Subjects

The total number of fee-for-service Medicare beneficiaries meeting eligibility criteria in the 20% sample was 4,605,644, and 1,412,257 (30.7%) were excluded because they did not have at least 4 ambulatory evaluation and management visits in 2011 ( Appendix E1 , available online at http://www.annemergmed.com ). The excluded group of beneficiaries appeared to be healthier because they tended to be slightly younger, with a lower median HCC score and fewer ED episodes between 2011 and 2013.

The final study population included 3,193,387 fee-for-service Medicare beneficiaries. Almost 60% of the study population experienced an ED episode between 2011 and 2013 ( Table 1 ). Patients with an ED episode were more often women, older, black or Hispanic, Medicaid dual eligible, originally enrolled in Medicare as disabled, and sicker, as measured both by higher HCC scores and more visits and hospitalizations. The majority of ED episodes were ED visits alone (54.8%), followed by hospital admissions through the ED (32.1%) and then observation stays through the ED (13.0%) ( Table 2 ). A higher proportion of women had an ED visit and discharge than an observation stay or admission. Patients with ED visits alone also had lower illness burden according to HCC scores compared with those with an observation stay or admission. Patients who were admitted tended to be older and were more likely to be Medicaid dual eligible or disabled.

Related posts:

Adolescent Male With Right Shoulder Pain After Football Injury Association Between the Opening of Retail Clinics and Low-Acuity Emergency Department Visits Bending the Curve of Health Trajectories for Older Adults Discharged From the Emergency Department Yet Another Sedating the Agitated Patient: A Moving Target? Classified 2017 Advertising Rates & Information