Exploring the effects of transfers and readmissions on trends in population counts of hospital admissions for coronary heart disease: a Western Australian data linkage study.

OBJECTIVES: To develop a method for categorising coronary heart disease (CHD) subtype in linked data accounting for different CHD diagnoses across records, and to compare hospital admission numbers and ratios of unlinked versus linked data for each CHD subtype over time, and across age groups and sex.
DESIGN: Cohort study. DATA SOURCE: Person-linked hospital administrative data covering all admissions for CHD in Western Australia from 1988 to 2013. MAIN OUTCOME: Ratios of (1) unlinked admission counts to contiguous admission (CA) counts (accounting for transfers), and (2) 28-day episode counts (accounting for transfers and readmissions) to CA counts stratified by CHD subtype, sex and age group.
RESULTS: In all CHD subtypes, the ratios changed in a linear or quadratic fashion over time and the coefficients of the trend term differed across CHD subtypes. Furthermore, for many CHD subtypes the ratios also differed by age group and sex. For example, in women aged 35-54 years, the ratio of unlinked to CA counts for non-ST elevation myocardial infarction admissions in 2000 was 1.10, and this increased in a linear fashion to 1.30 in 2013, representing an annual increase of 0.0148.
CONCLUSION: The use of unlinked counts in epidemiological estimates of CHD hospitalisations overestimates CHD counts. The CA and 28-day episode counts are more aligned with epidemiological studies of CHD. The degree of overestimation of counts using only unlinked counts varies in a complex manner with CHD subtype, time, sex and age group, and it is not possible to apply a simple correction factor to counts obtained from unlinked data. © Article author(s) (or their employer(s) unless otherwise stated in the text of the article) 2017. All rights reserved. No commercial use is permitted unless otherwise expressly granted.

Use of state-wide administrative data captures all hospital admissions in Western Australia.

Record linkage allowed the identification of contiguous admissions to account for transfers and 28-day episodes to account for readmissions.

While the complex pattern of counts and ratios presented is from a single jurisdiction in Australia, it is likely that the methods described will be generalisable to other states and territories. However, the ratios obtained may not be generalisable outside Western Australia (because of differences in healthcare systems) or beyond the study period.

Another limitation is the validity of coding for coronary heart disease in administrative data.

The use of 28-day episodes may miss a small number of related readmissions that occur beyond 28 days.

Introduction

Coronary heart disease (CHD) remains a major cause of death in Australia.1 Clinically it manifests across a spectrum of subtypes, from ST elevation myocardial infarction (STEMI) (the most severe), non-STEMI (NSTEMI), unstable angina and stable angina, through to other chronic presentations. There is increasing evidence that less severe forms of CHD, such as stable angina, also have an increased risk of major adverse cardiovascular events.2 Therefore, accurate information on population trends in CHD event rates and its subtypes is an indicator of the healthcare burden and essential for planning and evaluation of appropriate public health measures and clinical services. The focus on myocardial infarction (MI) alone fails to provide a complete understanding of the size of the problem of suspected CHD or its outcomes, and reliable estimates of CHD events at the population level are predicated on accurate stratification of CHD subtypes, for which there are limited data in Australia.

Population hospital administrative data provide a valuable data source in this regard where each admission is a separate record and diagnosis. However, this data source is not specifically designed for research purposes, and admission counts are susceptible to overinflation if the patient is transferred or readmitted multiple times during their clinical course for essentially a single episode of care. Additionally, recording of CHD subtype can differ between records in the same episode of care, requiring consideration when categorising CHD subtype for the episode. This is especially true for the management of CHD, which has historically been characterised with high rates of hospital transfers and early readmissions.3 Indeed, contemporary Australian data have shown that around 18%–30% of patients hospitalised for MI are transferred to another hospital,4 5 often for highly specialised coronary artery procedures, most notably coronary angiography and revascularisation by percutaneous coronary intervention (PCI). These specialised coronary care services are generally located at major population centres, and many patients, especially those from non-urban areas, are transferred to one of these hospitals for treatment and management of their condition.6 In addition, a significant number of patients with MI are readmitted for complications post-MI (such as repeat MI or heart failure), for elective procedures (such as coronary artery revascularisation or electrophysiological investigation), and to a lesser degree, for non-cardiac-related admissions.7 8

There is a potential to overestimate hospitalisation rates of CHD subtypes using unlinked data because transfers and readmissions are not accounted for. This could differentially affect CHD subtype rates, depending on the use of different diagnosis codes when patients are transferred or for early readmissions. For jurisdictions where only unlinked data are available, it is important to understand the degree of overestimation of the number of admissions across subtypes, and whether this changes over time or by age group and sex. Where person-linked hospital data are available, there is a need to assign a single relevant diagnosis to a group of admissions related by transfers or readmissions. To the best of our knowledge, approaches to these issues have not been addressed previously. Hence, our aims were to (1) develop an approach to identify and categorise admissions for each CHD subtype accounting for different CHD diagnoses across hospital transfers and readmission records from linked hospital data; (2) compare counts of unlinked CHD admissions with linked data accounting for transfers and readmissions; and (3) examine whether the ratios of these counts show similar or disparate patterns over time and across age and sex groups for each CHD subtype.

Methods

Data source and study population

For this cohort study, we used person-linked administrative health data from the Hospital Morbidity Data Collection, one of the core data sets of the Western Australian Data Linkage System. Western Australia (WA) is representative of national sociodemographic and health indicators,9 with an estimated resident population of 2.6 million in 2013.10 The available data set included all hospital records for any person hospitalised with CHD in WA from 1988 to 2013. We included all fatal and non-fatal admissions, with age restricted to 35–84 years. Variables available included demographic information, admission and discharge dates, principal and 20 secondary discharge diagnosis fields, and hospital locations.

Identification of CHD subtypes for individual (unlinked) admissions

All CHD admissions were identified from the principal discharge diagnosis field based on the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) (1 January 1988 to 30 June 1999) and the International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, Australian Modification (ICD-10-AM) (1 July 1999 to the present). CHD subtypes were defined as transmural MI/STEMI (ICD-9-CM: 410.0–410.6, 410.8; ICD-10-AM: I21.0-I21.3) (hereafter STEMI), subendocardial MI/NSTEMI (410.7; I21.4) (hereafter NSTEMI), unspecified MI (410.9; I21.9), unstable angina (411.1; I20.0), stable angina (413; I20.1-I20.9), and Other CHD (411.0, 411.81, 411.89, 412, 414; I23-I25). Other CHD includes complications following MI and chronic ischaemic heart disease. An addition to the labelling of transmural or subendocardial MI was added in ICD-10-AM in 2004, with reference to STEMI (‘transmural or STEMI’) and NSTEMI (‘subendocardial or NSTEMI’) included. All MI is a combination of STEMI, NSTEMI and unspecified MI; acute coronary syndrome (ACS) is a combination of All MI and unstable angina; and All CHD is a combination of ACS, stable angina and Other CHD.

Identifying transfers and readmissions

An interhospital transfer occurs when a patient is discharged from one hospital and directly admitted to another hospital within 1 day. Patients can have multiple transfers related to the same presentation. We introduce the concept of a contiguous admission (CA), which may represent a single isolated admission or an uninterrupted continuous hospital stay as a result of one or more transfers between hospitals. The admission date for the CA is the admission date of the first admission in the sequence. We also define a 28-day episode of care, which comprises an index CA and any subsequent CAs occurring within 28 days of the admission date of the index CA. A CA that begins more than 28 days after the index CA is considered a new episode of care. The 28-day period is commonly used in epidemiological studies.11–13

Assigning principal diagnosis for CHD subtype to each CA and 28-day episode

Each admission in a CA has its own principal discharge diagnosis code that may vary between admissions. We have calculated CA counts based on four approaches described below.

Diagnosis hierarchy

This is based on the work of Sanfilippo et al,14 and reflects the severity of the CHD subtypes from STEMI (most severe), NSTEMI, unstable angina and stable angina, to Other CHD (least severe). For a CA with multiple principal diagnoses, the most severe diagnostic category is used.

Hospital hierarchy

The hierarchy is metropolitan tertiary hospital (specialised cardiac care, diagnostic angiography and PCI), private metropolitan hospital (with and without aforementioned tertiary care), metropolitan non-tertiary hospital and rural/remote hospital. During the study period, all three metropolitan tertiary and four private hospitals had a cardiac catheter laboratory.15 None of the metropolitan non-tertiary or rural/remote hospitals had a cardiac catheter laboratory at the time of this study. For a CA with multiple principal diagnoses, the principal diagnosis from the hospital highest in the hierarchy is used.

First admission

The principal diagnosis recorded from the first admission in the CA is used. Given the acute nature of CHD, the first admission in a CA is presumed to be due to this condition, while subsequent transfers are for procedures or resultant complications or cardiac rehabilitation.

Last admission

The principal diagnosis recorded from the last admission in the CA is used. The last hospital admission in the CA is presumed to be when the most definitive diagnosis is made among all admissions.

The diagnostic CHD subtype assigned to each 28-day episode was based on the diagnosis hierarchy approach, that is, the most severe subtype of all the CAs that comprise the 28-day episode is used. Table 1 illustrates how diagnoses (CHD subtypes) are assigned to CAs (four approaches) and to 28-day episodes for a hypothetical patient with 10 hospital admissions, grouped into four CAs and three 28-day episodes.

Table 1

Example of typical patient record (not a real patient) depicting the different approaches of assigning a diagnosis to CA and 28-day episodes

Record no.	Patient ID	Admission date	Discharge date	Hospital	Transfer	28-day readmission	CA no.	28-day episode no.	Principal diagnosis	Diagnosis approach at the CA level				Diagnosis at 28-day episode level
										Diagnosis hierarchy	Hospital hierarchy	First admission	Last admission
1	1	1 February 2005	2 February 2005	Rural			1	1	Stable angina	NSTEMI	Unstable angina	Stable angina	NSTEMI	STEMI
2	1	2 February 2005	4 February 2005	Metropolitan tertiary	1		1	1	Unstable angina
3	1	4 February 2005	6 February 2005	Metropolitan non-tertiary	1		1	1	NSTEMI
4	1	17 February 2005	18 February 2005	Metropolitan tertiary		1	2	1	STEMI	STEMI	STEMI	STEMI	Other CHD	–
5	1	18 February 2005	22 February 2005	Private	1	1	2	1	Other CHD
6	1	10 October 2005	11 October 2005	Metropolitan non-tertiary			3	2	Stable angina	NSTEMI	NSTEMI	Stable angina	NSTEMI	NSTEMI
7	1	11 October 2005	14 October 2005	Metropolitan tertiary	1		3	2	NSTEMI
8	1	1 December 2005	2 December 2005	Rural			4	3	Non-CHD	Stable angina	Stable angina	Non-CHD	Non-CHD	Stable angina
9	1	2 December 2005	3 December 2005	Metropolitan tertiary	1		4	3	Stable angina
10	1	3 December 2005	5 December 2005	Private	1		4	3	Non-CHD

CA, contiguous admission; CHD, coronary heart disease; NSTEMI, non-STEMI; STEMI, ST elevation myocardial infarction.

Statistical analysis

Annual counts for each CHD subtype and combination subtypes are presented at the unlinked, CA and 28-day episode levels for 1988–2013. The ratio of unlinked admission count to CA count was calculated for each age group (35–54 years, 55–74 years, 75–84 years) and gender in each year to determine the relative overestimation of each CHD subtype. To examine the impact on counts from using 28-day episodes, we calculated the ratio of 28-day episode to CA counts for each age group and gender in each year, for each CHD subtype. Linear regression (with robust SEs) was used to compare the ratios statistically across age groups and gender, and assess trends over time. This analysis was restricted to the period 2000–2013 as CHD counts were more consistent during this time. All models included sex, age group, sex × age group interaction term and year as a continuous variable, and year squared was also included where a curved trend was indicated (Wald test, P<0.01). We fitted extended models with time interaction terms to test if there were differences in time trends by sex and age group (ie, we tested sex × year, age group × year, and sex × age group × year for ratios without curved trends, and for ratios with curved trends, also tested sex × year squared, age group × year squared, and sex × age group × year squared). Only a few of the time interaction tests had P<0.01, and in lieu of the large number of time interactions tested and the lack of any consistent pattern to these results, these were considered not to be real and were ignored (ie, considered as false-positive time interactions). Analyses were performed using Stata V.13.1.

Ethics approval

The study was granted a waiver of informed consent.

Results

There were 296 659 unlinked hospital admissions for CHD from 1988 to 2013 in WA (table 2). The diagnosis hierarchy approach resulted in the highest count of CHD admissions (n=273 793), and the approach based on the diagnosis from last admission resulted in the lowest count (n=263 313). The number of 28-day episodes was 242 966. The counts at the unlinked, CA level and 28-day episode level for each CHD subtype are shown in table 2.

Table 2

Diagnosis counts at the unlinked, CA and 28-day episode levels for admission years 1988–2013

	Unlinked level	CA level				28-day episode level
		Diagnosis hierarchy	Hospital hierarchy	Diagnosis based on first admission	Diagnosis based on last admission
CHD records (n)	296 659	273 793	269 614	267 389	263 313	242 966
Diagnosis
STEMI	37 457 (12.63%)	34 435 (12.58%)	33 313 (12.36%)	32 165 (12.03%)	32 014 (12.16%)	33 364 (13.73%)
NSTEMI	29 203 (9.84%)	24 734 (9.03%)	23 956 (8.89%)	21 868 (8.18%)	22 631 (8.59%)	23 738 (9.77%)
Unstable angina	72 223 (24.35%)	65 589 (23.96%)	63 301 (23.48%)	64 478 (24.11%)	60 333 (22.91%)	59 144 (24.34%)
Stable angina	77 076 (25.98%)	73 994 (27.03%)	73 898 (27.41%)	73 845 (27.62%)	73 037 (27.74%)	64 669 (26.62%)
Other CHD	69 070 (23.27%)	65 161 (23.80%)	65 751 (24.39%)	64 632 (24.17%)	66 148 (25.11%)	52 688 (21.68%)
All MI	78 315 (26.40%)	69 049 (25.22%)	66 664 (24.73%)	64 434 (24.10%)	63 818 (24.24%)	66 487 (27.36%)
ACS	150 538 (50.74%)	134 638 (49.18%)	129 965 (48.20%)	128 912 (48.21%)	124 151 (47.15%)	125 631 (51.71%)

ACS, acute coronary syndrome; CA, contiguous admission; CHD, coronary heart disease; MI, myocardial infarction; NSTEMI, non-STEMI; STEMI, ST elevation myocardial infarction.

Figure 1 shows trends in annual admission counts for each CHD subtype and combination subtypes at the CA level, using the diagnosis hierarchy approach and the three alternative approaches. The diagnosis hierarchy approach resulted in highest counts for the more severe CHD subtypes compared with the three alternative approaches, but all methods had similar trends over time for each CHD subtype.

Figure 1

Comparison of CHD counts from 1988 to 2013 using four different approaches at the contiguous admission level. ACS, acute coronary syndrome; CHD, coronary heart disease; MI, myocardial infarction; NSTEMI, non-STEMI; STEMI, ST elevation myocardial infraction.

Figure 2 compares annual CHD counts at the unlinked, CA (using diagnosis hierarchy approach) and 28-day episode levels from 1988 to 2013. The use of unlinked records resulted in the highest counts of all subtypes, while 28-day episode records resulted in the lowest counts. The difference between unlinked and CA counts tended to be greater in the latter half of the study period for STEMI, NSTEMI and unstable angina, while the reverse was apparent for Other CHD. The difference between unlinked and CA counts for NSTEMI, All MI and ACS increased from around 2000 onwards. The difference between CA and 28-day episode counts tended to increase from around 2000 onwards for NSTEMI but narrowed for STEMI and unstable angina.

Figure 2

Comparison of CHD counts at the unlinked, CA (diagnosis hierarchy approach) and 28-day episode levels from 1988 to 2013. ACS, acute coronary syndrome; CA, contiguous admission; CHD, coronary heart disease; MI, myocardial infarction; NSTEMI, non-STEMI; STEMI, ST elevation myocardial infarction.

Table 3 and online supplementary table 1 present estimated ratios for unlinked to CA counts (based on diagnosis hierarchy approach) from fitted regression models by CHD subtype, sex and age group for the period 2000–2013. In women aged 35–54 years, the ratio of unlinked to CA counts for NSTEMI admissions in 2000 was 1.10 (ie, 10% higher in unlinked), and this increased in a linear fashion to 1.30 (ie, 30% higher) in 2013, representing an increase of 0.0148 per year. Conversely, the overcount for STEMI and All MI followed a curved (quadratic) trend. For subtypes with a linear trend, the trend coefficients were largest in the most severe CHD subtype (NSTEMI: increase of 0.0148/year) and smallest in the least severe subtype (Other CHD: non-significant increase of 0.0003/year). The sex × age group interaction term was not significant in any individual or combination subtype, but the ratios were significantly higher in the youngest age group for STEMI, NSTEMI, stable angina and all combination subtypes. Men had significantly higher ratios than women for unstable angina and ACS.

Table 3

Estimated ratios of unlinked versus CA-level counts (diagnosis hierarchy approach) from 2000 to 2013, by CHD subtype, sex and age group

Diagnosis	Sex	Age group	Ratio (CI) in 2000	Ratio (CI) in 2013	Trend coefficient (P value)		P values for tests comparing
					Year	Year squared	Sex*	Age group*	Sex × age group
STEMI†	F	35–54	1.12 (1.09 to 1.15)	1.14 (1.11 to 1.18)	0.0174 (0.000)	−0.0012 (0.000)
	F	55–74	1.09 (1.06 to 1.11)	1.11 (1.08 to 1.14)	0.0174 (0.000)	−0.0012 (0.000)
	F	75–84	1.05 (1.02 to 1.09)	1.07 (1.02 to 1.13)	0.0174 (0.000)	−0.0012 (0.000)	0.946	0.000	0.945
	M	35–54	1.12 (1.10 to 1.15)	1.15 (1.11 to 1.19)	0.0174 (0.000)	−0.0012 (0.000)
	M	55–74	1.08 (1.06 to 1.10)	1.10 (1.08 to 1.13)	0.0174 (0.000)	−0.0012 (0.000)
	M	75–84	1.05 (1.03 to 1.08)	1.08 (1.04 to 1.11)	0.0174 (0.000)	−0.0012 (0.000)
NSTEMI‡	F	35–54	1.10 (1.08 to 1.13)	1.30 (1.27 to 1.33)	0.0148 (0.000)
	F	55–74	1.08 (1.06 to 1.10)	1.27 (1.25 to 1.29)	0.0148 (0.000)
	F	75–84	1.04 (1.02 to 1.06)	1.23 (1.21 to 1.25)	0.0148 (0.000)		0.768	0.000	0.734
	M	35–54	1.12 (1.09 to 1.14)	1.31 (1.29 to 1.33)	0.0148 (0.000)
	M	55–74	1.07 (1.06 to 1.09)	1.27 (1.25 to 1.28)	0.0148 (0.000)
	M	75–84	1.03 (1.02 to 1.05)	1.23 (1.21 to 1.24)	0.0148 (0.000)
Unstable angina‡	F	35–54	1.09 (1.07 to 1.10)	1.15 (1.13 to 1.16)	0.0047 (0.000)
	F	55–74	1.09 (1.07 to 1.10)	1.15 (1.14 to 1.16)	0.0047 (0.000)
	F	75–84	1.07 (1.06 to 1.09)	1.14 (1.12 to 1.15)	0.0047 (0.000)		0.000	0.010	0.445
	M	35–54	1.12 (1.10 to 1.13)	1.18 (1.16 to 1.20)	0.0047 (0.000)
	M	55–74	1.10 (1.09 to 1.11)	1.16 (1.15 to 1.17)	0.0047 (0.000)
	M	75–84	1.09 (1.08 to 1.11)	1.15 (1.14 to 1.17)	0.0047 (0.000)
Stable angina‡	F	35–54	1.05 (1.04 to 1.06)	1.07 (1.06 to 1.08)	0.0015 (0.000)
	F	55–74	1.03 (1.03 to 1.04)	1.05 (1.05 to 1.06)	0.0015 (0.000)
	F	75–84	1.03 (1.03 to 1.04)	1.05 (1.05 to 1.06)	0.0015 (0.000)		0.031	0.000	0.776
	M	35–54	1.04 (1.04 to 1.05)	1.06 (1.05 to 1.07)	0.0015 (0.000)
	M	55–74	1.03 (1.03 to 1.03)	1.05 (1.04 to 1.05)	0.0015 (0.000)
	M	75–84	1.03 (1.02 to 1.04)	1.05 (1.04 to 1.06)	0.0015 (0.000)
Other CHD‡	F	35–54	1.05 (1.03 to 1.07)	1.06 (1.03 to 1.08)	0.0003 (0.583)
	F	55–74	1.03 (1.02 to 1.04)	1.04 (1.03 to 1.05)	0.0003 (0.583)
	F	75–84	1.04 (1.03 to 1.06)	1.05 (1.03 to 1.06)	0.0003 (0.583)		0.057	0.072	0.630
	M	35–54	1.04 (1.03 to 1.05)	1.04 (1.03 to 1.05)	0.0003 (0.583)
	M	55–74	1.03 (1.02 to 1.04)	1.03 (1.03 to 1.04)	0.0003 (0.583)
	M	75–84	1.03 (1.02 to 1.04)	1.04 (1.03 to 1.05)	0.0003 (0.583)

*P values for sex and age group are from the respective models but without the age group × sex interaction term.

†From the model: ratio=constant + (admission year – 2000) + age group + sex + age group × sex + (admission year – 2000)2.

‡From the model: ratio=constant + (admission year – 2000) + age group + sex + age group × sex.

CA, contiguous admission; CHD, coronary heart disease; F, female; M, male;NSTEMI, non-STEMI; STEMI, ST elevation myocardial infarction.

Table 4 and online supplementary table 2 present the estimated ratios for CA versus 28-day episode counts. For example, in women aged 35–54 years, the ratio for STEMI was 1.10 in 2000 (ie, 10% higher for CA counts), and this decreased to 1.01 in 2013 (ie, 1% higher), representing a 0.0064 decrease per year. Ratios for unstable angina, stable angina, Other CHD and All CHD followed a curved (quadratic) trend. For example in women aged 35–54 years, the ratio for unstable angina was 1.15 in 2000 before levelling out at 1.09 from 2010 onwards. For unstable angina, stable angina, Other CHD, ACS and All CHD, the ratios were significantly higher in men than in women. Differences in ratios between age groups were seen for All CHD subtypes except for NSTEMI and Other CHD.

Table 4

Estimated ratios of CA versus 28-day episode-level counts (diagnosis hierarchy approach) from 2000 to 2013, by CHD subtype, sex and age group

Diagnosis	Sex	Age group	Ratio (CI) in 2000	Ratio (CI) in 2013	Trend coefficient (P value)		P values for tests comparing
					Year	Year squared	Sex*	Age group*	Sex × age group
STEMI†	F	35–54	1.10 (1.05 to 1.14)	1.01 (0.99 to 1.03)	−0.0064 (0.000)
	F	55–74	1.08 (1.06 to 1.09)	1.01 (1.00 to 1.02)	−0.0064 (0.000)
	F	75–84	1.05 (1.03 to 1.06)	1.03 (1.02 to 1.05)	−0.0064 (0.000)		0.946	0.000	0.799
	M	35–54	1.09 (1.06 to 1.11)	1.01 (0.98 to 1.03)	−0.0064 (0.000)
	M	55–74	1.08 (1.07 to 1.09)	1.01 (1.00 to 1.02)	−0.0064 (0.000)
	M	75–84	1.05 (1.04 to 1.06)	1.04 (1.02 to 1.06)	−0.0064 (0.000)
NSTEMI‡	F	35–54	1.10 (1.06 to 1.14)	1.05 (1.03 to 1.08)	−0.0036 (0.000)
	F	55–74	1.07 (1.06 to 1.09)	1.03 (1.01 to 1.04)	−0.0036 (0.000)
	F	75–84	1.08 (1.06 to 1.10)	1.04 (1.02 to 1.05)	−0.0036 (0.000)		0.470	0.937	0.033
	M	35–54	1.08 (1.06 to 1.10)	1.03 (1.02 to 1.05)	−0.0036 (0.000)
	M	55–74	1.10 (1.08 to 1.12)	1.05 (1.04 to 1.07)	−0.0036 (0.000)
	M	75–84	1.09 (1.07 to 1.11)	1.04 (1.03 to 1.06)	−0.0036 (0.000)
Unstable angina§	F	35–54	1.15 (1.13 to 1.17)	1.09 (1.07 to 1.10)	−0.0124 (0.000)	0.0006 (0.001)
	F	55–74	1.15 (1.14 to 1.17)	1.09 (1.08 to 1.10)	−0.0124 (0.000)	0.0006 (0.001)
	F	75–84	1.17 (1.15 to 1.19)	1.11 (1.09 to 1.12)	−0.0124 (0.000)	0.0006 (0.001)	0.000	0.040	0.017
	M	35–54	1.18 (1.16 to 1.20)	1.12 (1.10 to 1.14)	−0.0124 (0.000)	0.0006 (0.001)
	M	55–74	1.21 (1.19 to 1.22)	1.15 (1.13 to 1.16)	−0.0124 (0.000)	0.0006 (0.001)
	M	75–84	1.20 (1.18 to 1.21)	1.13 (1.12 to 1.15)	−0.0124 (0.000)	0.0006 (0.001)
Stable angina§	F	35–54	1.14 (1.12 to 1.16)	1.08 (1.07 to 1.10)	−0.0115 (0.000)	0.0006 (0.004)
	F	55–74	1.15 (1.14 to 1.17)	1.10 (1.09 to 1.11)	−0.0115 (0.000)	0.0006 (0.004)
	F	75–84	1.14 (1.12 to 1.16)	1.08 (1.07 to 1.10)	−0.0115 (0.000)	0.0006 (0.004)	0.000	0.009	0.097
	M	35–54	1.20 (1.18 to 1.23)	1.15 (1.13 to 1.16)	−0.0115 (0.000)	0.0006 (0.004)
	M	55–74	1.20 (1.18 to 1.22)	1.14 (1.13 to 1.15)	−0.0115 (0.000)	0.0006 (0.004)
	M	75–84	1.18 (1.16 to 1.20)	1.13 (1.11 to 1.14)	−0.0115 (0.000)	0.0006 (0.004)
Other CHD§	F	35–54	1.18 (1.15 to 1.22)	1.11 (1.08 to 1.14)	−0.0211 (0.000)	0.0012 (0.000)
	F	55–74	1.19 (1.16 to 1.22)	1.12 (1.10 to 1.14)	−0.0211 (0.000)	0.0012 (0.000)
	F	75–84	1.19 (1.16 to 1.21)	1.11 (1.09 to 1.13)	−0.0211 (0.000)	0.0012 (0.000)	0.000	0.359	0.542
	M	35–54	1.22 (1.20 to 1.25)	1.15 (1.13 to 1.17)	−0.0211 (0.000)	0.0012 (0.000)
	M	55–74	1.23 (1.20 to 1.25)	1.15 (1.14 to 1.17)	−0.0211 (0.000)	0.0012 (0.000)
	M	75–84	1.21 (1.19 to 1.24)	1.14 (1.12 to 1.15)	−0.0211 (0.000)	0.0012 (0.000)

*P values for sex and age group are from the respective models but without the age group × sex interaction term.

†From the model: ratio=constant + (admission year – 2000) + age group + sex + age group × sex + age × (admission year – 2000).

‡From the model: ratio=constant + (admission year – 2000) + age group + sex + age group × sex.

§From the model: ratio=constant + (admission year – 2000) + age group + sex + age group × sex + (admission year – 2000)2.

CA, contiguous admission; CHD, coronary heart disease; F, female; M, male; NSTEMI, non-STEMI; STEMI, ST elevation myocardial infarction.

Discussion

We developed different approaches to assign CHD diagnoses to a sequence of consecutive admissions and 28-day episodes that account for transfers and readmissions, thereby avoiding the overcount that occurs with unlinked administrative data. Hospitalisation data from 1988 to 2013 show that for each CHD subtype, unlinked records overcounted the number of CHD hospitalisations relative to CA counts and 28-day episode counts. Our analyses of ratios from 2000 to 2013 showed a complex pattern of overcounting in unlinked data due to transfers and readmissions. In almost All CHD subtypes, the ratios changed in a linear or quadratic fashion over time, and the coefficients of the trends differed across CHD subtypes. Further, for many CHD subtypes, the ratios also differed by age group and sex.

The development of the CA method accounts for transfers and allows for classification by CHD subtype where multiple admissions with differing discharge diagnoses are present. As each transfer and admission to the receiving hospital has its own principal discharge diagnosis, we compared four approaches to assigning a single clinically relevant diagnosis for each CA. Of the four approaches to assigning diagnosis, we contend that diagnosis hierarchy is the most clinically relevant approach and indicator of healthcare burden as it prioritises disease severity according to a physician’s clinical judgement. Of the four approaches, diagnosis hierarchy results in the highest CHD counts and would therefore result in the most conservative differences between unlinked and CA. Hospital hierarchy is based on resourcing of hospitals with coronary care services and the level of resourcing may differ in other jurisdictions. The recent introduction of coronary care services in rural hospitals in WA means that the hospital hierarchy method may become less applicable. Diagnosis based on first or last admission in a CA may not identify CHD-related admissions that occur in the middle of a CA, highlighted by the resulting low counts that occurred when using these methods to assign a diagnosis. A small number of patients have an MI during an admission for non-cardiac conditions16 and diagnosis based on first admission may not identify these CHD cases if they are subsequently transferred.

The ratios of unlinked versus CA counts for almost all subtypes (except STEMI and All MI) increased in a linear fashion, indicating a consistent increase in the overinflation of admission numbers in unlinked data due to transfers. This likely reflects a complex mix of changes in clinical guidelines and practice, facilitated by direct transfers to hospitals with PCI capability for ACS cases and prehospital care protocols during this period. The widening difference between unlinked and CA counts for NSTEMI indicates an increasing rate of transfer for this group of patients. Given that patients with NSTEMI are still at risk of future adverse events, clinical guidelines now recommend that these patients undergo early coronary angiography and hospitalisation if indicated.2 17 Patients who are not at a hospital with advanced coronary care services may be transferred as a priority to a hospital with such capabilities. These findings show that the use of unlinked data would bias temporal trends in NSTEMI hospitalisation rates upwards and that linked data, using the described methods, would provide more reliable trend estimates for hospitalisation rates of NSTEMI in particular.

Furthermore, ratios were higher in the younger than older age groups for all subtypes, indicating that older patients with CHD were less likely to be transferred than younger patients. We also found men had a higher ratio than women for unstable angina and ACS. These sex and age differences in transfers may partly reflect age and sex disparities in ACS care and especially invasive management reported in earlier studies,18 19 although further studies are needed to support this theory.

Twenty-eight-day episodes have previously only been used to capture early MI readmissions following an index MI admission, thus reducing overestimation of population rates for MI. Historically, early readmissions were often for coronary procedures or other management related to the initial MI admission. Our method ensures 28-day episodes capture any CHD readmission during this period. In general, our results show that early readmissions across All CHD subtypes have decreased, although the trend was not linear for unstable and stable angina, and Other CHD. This could indicate that most acute treatment is now managed during the initial admission or subsequent transfer, thus requiring fewer readmissions.

The findings of this study have important implications for monitoring population trends in MI and Other CHD subtypes. The ratios of counts we presented would have been the same if we had used age-standardised rates (ASRs) as population denominators would have been the same in all three levels of counts. The trends in CA and 28-day episode counts for STEMI and NSTEMI are in accordance with other studies showing that hospital admissions for STEMI have decreased in Western countries, while admissions for NSTEMI have increased.20 21 The use of the CA and 28-day episode methods in linked data offsets overcounting of MI events, which could potentially inflate trends in ASRs. The effect of overestimation of MI hospitalisation numbers due to transfers and readmissions could also artificially reduce case fatality because of the impact on case fatality denominators. In addition, it allows accurate representation of other subtypes of CHD, for which there are limited data at a whole-population level.

There are a number of jurisdictions including Australia where linked data are not available at a national/population level, for example, the USA, where studies reporting nationwide trends on MI or CHD rely on unlinked data (eg, Nationwide Inpatient Sample), or where the more recent introduction of national linked data necessitates use of unlinked data where long-term trends are required (eg, Hospital Episode Statistics data in England).22 23 Therefore we contend our methods and data will be of interest to countries outside of Australia. Although we have described an approach to dealing with transfers and defining episodes of care for use with CHD, these methods could be applied to other conditions that have high rates of transfer and readmissions, such as major trauma and head injury, where many patients are transferred from rural sites to major tertiary hospitals with intensive care and/or head injury units and rehabilitation.

Strengths and limitations

Strengths of our study include use of state-wide data that capture all hospital admissions in WA. Record linkage allowed the identification of CAs to account for transfers and 28-day episodes. The limitations of this study include the validity of coding for CHD. An earlier WA study using linked data showed that the sensitivity of hospital coding for MI was 76.9% in patients aged 35–69.14 The use of 28-day episodes may miss a small number of related readmissions that occur beyond 28 days. Furthermore, we did not adjust for confounders such as remoteness and Indigenous status, which may influence transfer and readmission patterns.6 24 The complex pattern of counts and ratios we presented is from WA for 2000–2013 and may not be generalisable to other jurisdictions (because of different healthcare systems) or beyond the study period; however, the methods we described are generalisable to other states and territories.

Conclusions

Although unlinked data have its place in measurement of hospital health service utilisation, its use in epidemiological estimates of CHD hospitalisations overestimates CHD counts. We contend that CA (accounting for transfers) and 28-day episode (accounting for transfers and readmissions) counts are more aligned with epidemiological studies of CHD. The degree of overestimation of counts using only unlinked records varies in a complex manner with CHD subtype, time, sex and age group, and it is not possible to apply a simple correction factor to counts obtained from unlinked data.