Reconciling hospital-acquired complications and CHADx+ in Victorian coded hospital data

Abstract

Background:

The Council of Australian Governments has focused the attention of health service managers and state health departments on a list of hospital-acquired complications (HACs) proposed as the basis of funding adjustments for poor quality of hospital inpatient care. These were devised for the Australian Commission on Safety and Quality in Health Care as a subset of their earlier classification of hospital-acquired complications (CHADx) and designed to be used by health services to monitor safety performance for their admitted patients.

Objective:

To improve uptake of both classification systems by clarifying their purposes and by reconciling the ICD-10-AM code sets used in HACs and the Victorian revisions to the CHADx system (CHADx+).

Method:

Frequency analysis of individual clinical codes with condition onset flag (COF 1) included in both classification systems using the Victorian Admitted Episodes Dataset for 2014/2015 (n = 2,623,275 separations). Narrative description of the resulting differences in definition of “adverse events” embodied in the two systems.

Results:

As expected, a high proportion of ICD-10-AM codes used in the HACs also appear in CHADx+, and given the wider scope of CHADx+, it uses a higher proportion of all COF 1 diagnoses than HACs (82% vs. 10%). This leads to differing estimates of rates of adverse events: 2.12% of cases for HACs and 11.13% for CHADx+. Most CHADx classes (70%) are not covered by the HAC system; discrepancies result from the exclusion from HACs of several major CHADx+ groups and from a narrower definition of detailed HAC classes compared with CHADx+. Case exclusion criteria in HACs (primarily mental health admissions) resulted in a very small proportion of discrepancies (0.13%) between systems.

Discussion:

Issues of purpose and focus of these two Australian systems, HACs for clinical governance and CHADx+ for local quality improvement, explain many of the differences between them, and their approach to preventability, and risk stratification.

Conclusion:

A clearer delineation between these two systems using routinely coded hospital data will assist funders, clinicians, quality improvement professionals and health information managers to understand discrepancies in case identification between them and support their different information needs.

Keywords

classification health information management International Classification of Diseases patient safety quality and safety hospital-acquired conditions clinical coding financial management hospital health classification medical classification health services administration activity-based funding

Background

The quality and safety of hospital care has become more prominent in healthcare policy discussions since the creation of the Australian Commission on Safety and Quality in Health Care (ACSQHC) in 2006 (ACSQHC, 2017a). Among its achievements has been the reform of the hospital accreditation process and adoption of nationally endorsed hospital quality standards (ACSQHC, 2012) by all state and territory governments. In 2008, the ACSQHC sponsored research to understand the relative costs of different forms of adverse events (defined as diagnoses that are not present on admission, but arise after a patient is hospitalised, and require additional treatment, monitoring and/or extended length of stay). These conditions can be identified through the assignment of a nationally defined condition onset flag (COF)) (AIHW/METeOR, 2017).

This definition of adverse events was used in the development of the classification of hospital-acquired diagnoses (CHADx), a computer programme that identifies poor patient outcomes in routinely coded hospital admitted-patient data (Jackson et al., 2009a; Michel et al., 2009). Using data from two Australian states, complications were found to add 17.3% to total hospital expenditures (Jackson et al., 2011). The project noted that current safety and quality improvement efforts focus on infrequent adverse events with high episode costs, but give less attention to events or diagnoses that are lower cost but more frequent, and thus result in high costs to the system as a whole.

The CHADx algorithms (and associated data cleaning programme (Jackson et al., 2009b)) have been used subsequently by health services to identify outcomes of hospital-acquired conditions (Trentino et al., 2013) and patients at risk of complications (Cromarty et al., 2014). At the aggregated “Major CHADx” level, they have been used to report on state-level differences in rates of clinical coded complications by the Australian Institute of Health and Welfare (AIHW, 2016). CHADx was used in the recent Victorian review of hospital quality and safety to illustrate the ubiquity of quality problems (Duckett et al., 2016). It is currently under redevelopment (“CHADx+”) by the Victorian Department of Health and Human Services to incorporate user feedback and thus support wider use in medical specialty and hospital-level quality improvement programmes. A draft version of CHADx+ was used for this study.

In 2012, the Independent Hospital Pricing Authority (IHPA) formed a Joint Working Party with the ACSQHC to consider whether and how the national activity-based funding system could support quality improvement across the country. This group sponsored a study of CHADx costs using national hospital activity and cost data (Pearse et al., 2013). Building on earlier work in Victoria (McNair et al., 2009), the Pearse et al. report recommended four alternatives for incorporation of information on complications into national activity-based funding systems.

The Joint Working Party also sponsored a literature review of funding approaches to quality improvement (Eagar et al., 2013) and a project to develop indicators on which funding incentives could be applied. Working with a panel of doctors and patient safety experts, various CHADx groups were modified to define “high priority complications” for this purpose based on the criteria of “preventability,” “patient impact,” “health service impact” and “clinical priority” (KPMG and ACSQHC, 2013). The draft high priority complications (later, “Hospital Acquired Complications” or HACs) were also used in a demonstration project in 15 health services, the report of which concluded that “monitoring and reporting on HACs at the hospital level can be used by clinicians to detect patient safety problems and develop clinical risk mitigation strategies…” (ACSQHC, 2016a).

The Council of Australian Governments (COAG) has taken up this work in proposing the HAC list as the basis of funding adjustments for poor quality of hospital inpatient care (COAG, 2016). At the time of writing, IHPA has not determined which or how many of the HACs might be used in this way in the national funding model, nor which of the COAG’s four options for funding adjustment might be adopted.

Given the two classification systems (HACs and CHADx+) have different, but overlapping histories, we sought to explore the overlap in the two classifications in light of their proposed uses in the Australian healthcare system. Reconciling the ICD-10-AM code sets used in the HAC and CHADx+ classification systems, and clarifying their potential applications, may assist in the uptake of both by health services, state/territory, and national bodies.

Method

Design

Frequency analysis of individual clinical codes included in both classification systems using the Victorian Admitted Episodes Dataset for 2014/2015. Narrative description of the resulting differences in definition of “adverse events” embodied in the two classification systems.

Sample

All care type, public and private hospital, same-day and multi-day inpatient admissions to Victorian hospitals (n = 2,623,275).

Research variables and measurement

HAC indicators (N = 38 classes within 16 overarching classes (ACSQHC, 2016b) and CHADx+ (N = 159 classes with 17 overarching major CHADx+ or MCHADx+ classes (Jackson, 2017). COF 1 for “a condition which arises during the episode of admitted patient care and would not have been present on admission” (AIHW, METeOR, 2017).

HACs are further qualified with transfusion procedure codes (ACHI, ACCD, 2015a) for HAC 4.1 Post-operative haemorrhage/haematoma requiring transfusion and/or return to theatre, ventilation codes for HAC 6.1 Respiratory failure including acute respiratory distress syndrome requiring ventilation and dialysis codes for HAC 8.1 Renal failure requiring haemodialysis or continuous veno-venous haemodialysis. Additional external cause diagnosis codes are required to define fall injuries (HACs 2.1–2.3), anastomotic leak (HAC 4.3) and drug-related respiratory complications (HAC 10.1).

Six HACs (4.1–5.1) look forward to the introduction of new National Minimum Data Set (NMDS) markers of unplanned return to theatre and unplanned intensive care unit admission; of these, two classes (4.5 Other surgical complications requiring unplanned return to theatre and 5.1 Unplanned admission to intensive care unit) are currently inactive awaiting introduction of these markers. In addition to universal exclusions (mental health admissions, same-day treatments for chemotherapy or dialysis, unqualified newborns and care types 9 (organ procurement – posthumous) and 10 (hospital boarder)), HAC 8.1 Renal failure requiring haemodialysis or continuous veno-venous haemodialysis excludes admissions for patients with chronic kidney disease, stages 4 and 5. HAC 15.1 (Third and fourth degree perineal laceration during delivery) is qualified by a definition of vaginal birth; HAC 16.1 Neonatal birth trauma relates only to newborns at term or ≥2000 g, without osteogenesis imperfecta or brachial plexus injuries, and not transferred-in. Both the perinatal classes relax the requirement for COF 1 flags, and the neonatal birth trauma class allows use of the principal as well as additional diagnoses.

The original CHADx specifications have been published elsewhere (ACSQHC, 2017b; Jackson et al., 2009a). The Victorian Department of Health and Human Services has sponsored re-specification of the diagnosis-related classes in CHADx+ to take account of user feedback on the original CHADx code set. Only falls (CHADx+ 3_01–3_04) and adverse drug events (CHADx+ 2_01–2_18) now require a linkage between the defining diagnosis code (type of injury or drug-related harm) and a specific external cause code (type of fall, class of drug responsible).

The close linkage in the previous version was found to perform poorly on national data (Pearse et al., 2013), and thus many of the sequencing rules have been relaxed in the current version. The revisions reassign many CHADx+ classes in order to consolidate infections into a comprehensive overarching MCHADx+, separately report different kinds of sepsis (CHADx+ 4_01–4_06) and distinguish major cardiac arrhythmias (CHADx+ 5_05) from less serious ones (CHADx+ 5_06). CHADx+ code sets are still under development for additional procedures (CHAPx) and for readmission-related complications (RR-CHADx).

Most analysis is undertaken at the ICD-10-AM code or “instance” level, recognising that an episode may have more than one complication instance coded. Analysis of “instances” of code use recognises that in 50.1% of cases, only a single CHADx+ is recorded, and for these single-complication episodes, the “instance” will equate to a count of separations. Analysis of separations rather than instances is useful to compute the rate (percent of separations) of any hospital-acquired harm for the two classifications (see Overview, below) but not for other comparisons of code use undertaken here. We also compare the two systems at the “class” level (38 HACs, 159 CHADx+), as each level provides information for different system stakeholders.

Results

Overview

Table 1 shows the degree of overlap between the two classifications. Of all COF 1 diagnoses, 10% were common to both classifications, and the comprehensive CHADx+ uses 81.9% of these flagged codes overall. Coding standards require redundant additional codes for the same event (external cause codes, any specific injury, activity, and place of occurrence, all with the same COF) and account for the 138,011 code instances excluded from both classifications. The rates of “any complication” in the episode were 2.12% for HACs and 11.13% for CHADx+ (n = 54,785, and 291,874, of 2.623 million episodes, respectively).

Table 1.

HAC and CHADx+ use of COF 1 diagnosis codes.^a

HAC instances	CHADx+ instances
HAC instances	With CHADx+	Without CHADx+	All
With HAC	76,945	33	76,978
Without HAC	549,405	137,978	687,383
All	626,350	138,011	764,361

HAC: hospital-acquired complication; CHADx: classification of hospital-acquired diagnoses; COF: condition onset flag.

^a N = 2,623,275 acute care inpatient episodes, Victoria, 2014–2015. Newborn diagnosis codes deemed as hospital-acquired complications regardless of the COF in both HACs and CHADx+. Maternal principal diagnosis codes are counted in CHADx+ for the birth episode; these diagnoses are not in scope for HACs.

Similarities

Table 2 lists the six HAC and CHADx+ classes that use the same code subset. In some cases, the HACs divide code sets between two classes, but encompass the same codes and case identification as CHADx+. Thus, two CHADx+ classes are split into finer HACs: CHADx+ 5_05 Ventricular Fibrillation/cardiac arrest identifies the same cases, but these are divided between HACs 14.2 Arrhythmias and 14.3 Cardiac Arrest. Similarly, the CHADx+ combines stage 3 and 4 pressure injuries (CHADx+ 8_02) but these are separately reported as three HACs: 1.1 (Stage III Ulcer), 1.2 (Stage IV Ulcer) and 1.3 (Unspecified). Stage 1 and 2 pressure injuries are not reported in the HAC classification.

Table 2.

HAC/CHADx+ classes that use the same code subset.^a

HAC	HACs descriptors	CHADx+	HADs descriptors	CHADx diagnoses	Count of diagnoses in both HAC and CHADx	HAC totals	Count of diagnoses omitted by HAC risk adjustment^b
4.2	Surgical wound dehiscence	1_08	Disruption of wound	1482	1482		4
10.2	Haemorrhagic disorder due to circulating anticoagulants	2_05	Coagulation defect due to drugs affecting blood constituents	1881	1881		5
3.6	MRO	4_07	Antibiotic-resistant infections	1526	1526		25
14.4	Acute coronary syndrome including unstable angina, STEMI, NSTEMI	5_01	AMI	1601	1601		5
7.1	Pulmonary embolism	5_02	PE	772	772		4
14.3	Cardiac arrest	5_05	Ventricular fibrillation/cardiac arrest	1199	867		9
14.2	Arrhythmias				332	1199	2
1.1	Stage III decubitus ulcer and pressure area	8_02	Pressure injury stages 3 and 4 (includes unspecified)	1693	238		0
1.2	Stage IV decubitus ulcer and pressure area				44		0
1.3	Unspecified decubitus ulcer and pressure area				1411	1693	11
15.1	Third and fourth degree perineal laceration during delivery	12_08	Third degree and fourth degree perineal laceration	1512	1512		0
			Total	11,666	11,666		65

HAC: hospital-acquired complication; CHADx: classification of hospital-acquired diagnoses; HADs: hospital-acquired diagnoses; MRO: multi-resistant organism; AMI: acute myocardial infarction; PE: pulmonary embolism.

^a N = 2,623,275 acute care inpatient episodes, Victoria, 2014–2015.

^bRisk adjustment factors for each HAC available at https://www.safetyandquality.gov.au/hospital-acquired-complications-specifications-v1-1-oct-2016/.

Differences in coverage

The most obvious difference between the two systems is their coverage. Table 3 lists the number of CHADx+ classes with no equivalent in the HAC indicators by MCHADx+. Coverage exclusions for HACs include all codes in four MCHADx+ groups (11 Early pregnancy, 14 Non-drug related coagulation disorders and anaemias, 16 Nervous system complications and 17 Other complications). In addition, ≥75% of CHADx+ classes in five MCHADx+ roll-up categories are omitted from HACS: Adverse drug events (MCHADx+2), Hospital-acquired skin conditions (8), Hospital-acquired mental health complications (10), Labour and delivery (12) and Neonatal (13). In all, some 2165 CHADx+ COF 1 and perinatal diagnosis codes are omitted from the HAC specifications, representing 448,735 instances in the Victorian data.

Table 3.

CHADx+ classes with no HAC equivalent.^a

Major CHADx+	HADs for:	Classes wholly omitted by HACs	N of classes in CHADx+	% of classes omitted by HAC
MCHADx+1	Intra and post-procedural complications	16	23	69.6
MCHADx+2	Adverse drug events	14	18	77.8
MCHADx+3	Accidental injuries	4	7	57.1
MCHADx+4	Hospital-acquired infections	5	20	25.0
MCHADx+5	Cardiovascular complications	5	12	41.7
MCHADx+6	Respiratory complications	4	7	57.1
MCHADx+7	Gastrointestinal complications	4	6	66.7
MCHADx+8	Skin conditions	3	4	75.0
MCHADx+9	Genitourinary complications	2	4	50.0
MCHADx+10	Hospital-acquired psychiatric states	6	7	85.7
MCHADx+11	Early pregnancy complications	3	3	100.0
MCHADx+12	Labour and delivery complications	13	14	92.9
MCHADx+13	Perinatal complications	8	10	80.0
MCHADx+14	Haematological complications	4	4	100.0
MCHADx+15	Metabolic complications	4	6	66.7
MCHADx+16	Nervous system complications	3	3	100.0
MCHADx+17	Other complications	11	11	100.0
	Total	109	159	68.6

HAC: hospital-acquired complication; CHADx: classification of hospital-acquired diagnoses; HADs: hospital-acquired diagnoses.

^a N = 2,623,275 acute care inpatient episodes, Victoria, 2014–2015.

Differences in specificity of CHADx+ and HAC classes and inbuilt HAC risk adjustment

Differences between the two systems arising from differing approaches to the specificity of classes in each are summarised in Table 4. This compares the frequency of code use by each classification, and for each HAC reports the number of code instances excluded by its inbuilt risk adjustment approach. All HACs limit coverage to specific patient groups (e.g. excluding premature newborns (<2000 g) and mental health admissions), and some include procedure qualifiers (e.g. renal failure only with dialysis). The mental health exclusions have a relatively large impact on Delirium and hospital-acquired Urinary Tract Infection CHADx+ classes (HACs 3.1 Urinary tract infection and 11.1 Delirium), but in total only 973 instances are excluded by these factors (65 in wholly overlapping classes and 908 in split classes – Tables 3 and 4).

Table 4.

Overlapping CHADx+/HAC classes.

HAC	HACs descriptors	CHADx+	HADs descriptors	Count of CHADx by subclass	CHADx total	Count of diagnoses in both HAC (risk adjusted) and CHADx+	HAC total prior to risk adjustment	HAC diagnoses as % of CHADx+ instances	Count of diagnoses omitted by HAC risk adjustment^a
Single HACs partially mapping to single CHADx+
4.1	Post-operative haemorrhage/haematoma requiring transfusion and/or return to theatre	1_04	Haemorrhage and haematoma	7942	7942	2136	2138	26.9%	2
4.4	Vascular graft failure	1_09	Complications of cardiac and vascular implants	7819	7819	38	38	0.5%	0
3.2	Surgical site infection	4_08	Wound infections	2361	2361	2032	2041	86.4%	9
3.8	GI Infections	4_12	Hospital-acquired C_Diff and other GI infections	7040	7040	981	997	14.2%	16
7.2	Deep venous thrombosis	5_09	Venous thrombosis/embolism	2527	2527	865	875	34.6%	10
8.1	Renal failure	9_01	Acute renal failure	9770	9770	366	366	3.7%	0
12.1	Urinary incontinence	9_03	Other complications and symptoms of the urinary system	10,281	10,281	1938	1974	19.2%	36
13.1	Malnutrition	15_03	Hospital-acquired nutrition deficiencies	3644	3644	2025	2048	56.2%	23
Single HACs mapping to multiple CHADx+
3.1	Urinary tract infection	4_16	Hospital-acquired urinary tract infection	9179		8768			321
		4_17	Maternal infections^b	^b	9179	82	9171	99.9%	0
3.3	Pneumonia	4_09	Hospital-acquired pneumonia	5507		5421			45
		4_10	Hospital-acquired influenza/bronchitis	1963	7470	1614	7168	96.0%	88
3.4	Blood stream infection	4_01	Severe sepsis and septic shock	829		755			0
		4_02	Sepsis dt Strep	229		229			0
		4_03	Sepsis dt Staph	358		329			0
		4_04	Puerperal sepsis	134		133			0
		4_05	Newborn sepsis	2095		1620			12
		4_06	SIRS and other sepsis	2705		2515			10
		4_20	Other hospital-acquired infections	3518	9868	518	6126	62.1%	5
4.3	Anastomotic leak	1_12	Complications of other implants	2610		31			0
		1_20	Post-procedural disorders: Digestive system	2047		118			0
		1_22	Post-procedural disorders: Genitourinary system	939	5596	30	179	3.2%	0
6.2	Aspiration pneumonia	1_19	Post-procedural disorders: Respiratory system	1408		12			0
		6_02	Aspiration pneumonia	1839	3247	1828	1851	57.0%	11
9.1	GI bleeding	7_05	Gastrointestinal (GI) bleeding	2796		2735			60
		7_06	Other digestive system disorders	5472	8268	32	2828	34.2%	1
10.3	Hypoglycaemia	15_04	Glycaemic disorders	1182		405			17
		2_16	Complications due to other drugs	8179	9361	23	445	4.8%	0
11.1	Delirium	10_03	Delirium	9318		9172			129
		2_08	Altered mental state due to opioids	1590		437			1
		2_12	Altered mental state due to anaesthesia	374	11,282	123	9863	87.4%	1
14.1	Heart failure and pulmonary oedema	5_06	Heart failure	2730		2723			7
		6_03	Pulmonary oedema, pneumothorax and pleural effusion	3762	6492	264	2995	46.1%	1
14.2	Arrhythmias	5_03	Major cardiac arrhythmias	3768		2877			18
		5_04	Conduction disturbances/abnormal heart beat	22,241	26,009	7203	10,135	39.0%	37
16.1	Neonatal birth trauma	13_02	Intracranial haemorrhage, hypoxia and other brain injuries	1230		44			0
		13_03	Other birth trauma	2318	3548	261	305	8.6%	0
Single CHADx+ mapping to multiple HACs
3.5	Central line/peripheral line BSI	4_11	Device/implant-related infections	2468		1698			16
3.7	Prosthetic associated infection	4_11	Device/implant-related infections		2468	662	2376	96.3%
6.1	Respiratory failure including acute respiratory distress syndrome requiring ventilation	6_01	Respiratory distress/failure/pulmonary collapse	9325		1346			1
10.1	Drug-related respiratory complications/depression				9325	166	1513	16.2%
Multiple to multiple mapping
2.2	Fracture neck of femur	3_01	Falls with fractured femur	172		160			5
2.1	Falls with fracture or intracranial injuries	3_02	Falls with intracranial injury	204		183			13
2.3	Other fractures	3_03	Fall with other fractures	494	870	446	820	94.3%	13
			Total	16,4367	16,4367	65,344	66,252		908

HAC: hospital-acquired complication; CHADx: classification of hospital-acquired diagnoses; HADs: hospital-acquired diagnoses; GI: gastrointestinal; SIRS: systemic inflammatory response syndrome; BSI: bloodstream infection.

^aRisk adjustment factors for each HAC available at https://www.safetyandquality.gov.au/hospital-acquired-complications-specifications-v1-1-oct-2016/.

^bCannot be added because urinary tract infections in maternity patients are double counted to allow for monitoring by both obstetricians (MCHADx+12) and infection control consultants (MCHADx+4).

Coincident identification of cases (or overlap) ranges from only half a percent between Complications of cardiac and vascular implants (CHADx+ 1_09) and the narrower HAC 4.4 Vascular graft failure, to over 90% for classes identified as Falls with specified cranial and fracture injuries, Device/implant-related infections and Hospital-acquired pneumonia. HAC classes do not capture the 5308 Other fall injuries identified by CHADx+ 3_04.

The narrow focus of HAC 4.4 Vascular graft failure results in significant undercounting of other Complications of cardiac and vascular implants (CHADx+ 1_09, with 7658 additional complications recorded). This is of particular interest, as these complications have been shown to incur high system-wide incremental costs (Jackson et al., 2011). Similarly, HAC 3.7 identifies Infections associated with prosthetics/implantable devices (n = 367), but not the mechanical complications (breakdown, displacement, leakage, etc.) of other implants and prostheses (n = 13,337 in CHADx+ 1_09–1_12).

Complex instances of overlapped coverage are apparent in some CHADx+/HAC comparisons. In the overlapping classes for Haemorrhage, for example, HAC 4.1 Post-operative haemorrhage/haematoma requiring transfusion and/or return to theatre is defined as specifically post-operative and requiring a transfusion. This identifies 2136 instances, in contrast to the additional 4791 instances which do not meet the transfusion procedure criteria, but are coded to T81.0 Haemorrhage and haematoma complicating a procedure in CHADx+ 1_04. In addition, CHADx+ 1_04 includes 1015 other haemorrhages coded with the less specific symptom code R58 Haemorrhage not elsewhere classified. These omissions account for the 5804 instance difference between the two classes.

A second example of complex overlap is the HAC related to Anastomotic leak (4.3) which includes four instances of ICD code N99.8 Other post-procedural disorders of genitourinary system. These are counted in CHADx+ 1_22 as Post-procedural genitourinary diagnoses; in addition, CHADx+ 1_22 includes 30 overlapping cases of HAC 8.1, Renal failure requiring haemodialysis or continuous veno-venous haemodialysis. If the intent of these classifications is to improve quality and safety outcomes for patients, consideration needs to be given to how quality and safety staff can reconcile these differences. This is especially important when the overlaps involve different clinical specialties.

Our analysis identified three diagnosis codes included in HACs but not in CHADx+: K25.9 (gastric ulcer), K27.9 (peptic ulcer) and K28.9 (gastrojejunal ulcer) assigned to HAC 9.1 Gastrointestinal (GI) bleeding. These codes are among a set of GI diagnosis codes that share fourth character designations distinguished by various manifestations, where .9 indicates “…without haemorrhage or perforation.” These accounted for 33 instances in the Victorian data where a HAC instance did not overlap with a CHADx+. On review, these have not been added to the CHADx+ identification of gastrointestinal bleeding (7.2).

Table 5 summarises the major reasons for discrepancies between HACs and CHADx+ instances in the Victorian data, accounting for nearly half of all code-based discrepancies. HACs 8.1 (Renal failure requiring haemodialysis or continuous veno-venous haemodialysis), 6.1 (Respiratory failure including acute respiratory distress syndrome requiring ventilation), 4.1 (Post-operative haemorrhage/haematoma requiring transfusion and/or return to theatre) and 3.2 (Surgical site infection) all reflect criteria related to surgical patients or procedures patients have undergone as a result of the complication itself. All others reflect the ACSQHC deliberations, choosing to limit focus to complications with greater importance, according to the criteria of “preventability,” “patient impact,” “health service impact” and “clinical priority,” or limited to more specific (not “other” or “unspecified”) ICD codes. Table 6 shows that overall, the largest source of HAC exclusions (82%) is the difference in coverage, with a very small proportion attributable to risk adjustment.

Table 5.

Reasons for omissions in overlapping HAC/CHADx+ classes.^a

CHADx+	HACs	Reason for exclusions in the HAC classification	Code instances omitted
9_01	8.1	HAC for Renal Failure defined as “requiring haemodialysis…” in patients with no stage 4/5 chronic kidney disease	9404
6_01	6.1	Respiratory failure/collapse limited to episodes with respiratory support/ventilation	7812
1_09	4.4	HAC Vascular Graft Failure identified as subset of CHADx complications of cardiac and vascular implants	7781
4_12	3.8	HAC omits A09.9 Gastroenteritis and colitis of unspecified origin	6043
1_04	4.1	Haemorrhage and haematoma limited to surgical episodes	5804
1_12, 1_20, 1_22	4.3	HAC definition of anastomotic leak (Y832) combines urinary and GI anastomoses but omits other post-procedural complications	5417
5_09	7.2	HAC excludes “other & unspecified” codes (I80.8 and I80.9) for thrombophlebitis	1652
15_03	13.1	HAC nutrition deficiencies limited to protein–energy malnutrition	1596
4_05	3.4	COF 1 on principal diagnosis of newborn sepsis (P360–363) not counted in HAC 3.4	475
4_08	3.2	HAC Wound infections limited to surgical site infections	320
		Total	46,304

HAC: hospital-acquired complication; CHADx: classification of hospital-acquired diagnoses; GI: gastrointestinal.

^a N = 2,623,275 acute care inpatient episodes, Victoria, 2014–2015.

Table 6.

Summary of reasons for discrepancies.^a

CHADx+	HACs	Reason for exclusions in the HAC classification	Code instances omitted
109 classes	38	Differences in coverage	448,735
11 classes	10	Top 10 reasons for discrepancies (Table 5)	46,304
–	–	All other code-based exclusions related to procedure qualifiers, severity/preventability criteria	53,392
35 classes	–	Risk adjustment exclusions in HACs^b	974
		Total	549,405

HAC: hospital-acquired complication; CHADx: classification of hospital-acquired diagnoses.

^a N = 2,623,275 acute care inpatient episodes, Victoria, 2014–2015.

^bRisk adjustment factors for each HAC available at https://www.safetyandquality.gov.au/hospital-acquired-complications-specifications-v1-1-oct-2016/.

Discussion

Our analysis highlights the similarities and differences between two Australian systems using routine hospital admitted-patient data: one designed for clinical governance (and potentially, for funding) and the other intended to support local quality improvement.

Funders and system managers may rely on an indicator system such as the HACs with narrow coverage to reassure the public that hospital care is not grossly harmful. As an overview for clinical governance of hospitals, this limited set of risk adjusted, high-impact “preventable” conditions (HACs) will be useful to track between sites and over time. By contrast, comprehensive classification systems such as CHADx+ will support more focused and site-specific investigation, to allow quality improvement to be tailored to local problems, and incidentally, to compensate for varying levels of specificity in documentation that may prevent a HAC being identified.

Most apparent is the difference in the definition of “adverse event” that is used. In the HACs, the intent is to limit attention to a subset of all hospital-acquired (COF 1) diagnoses that have been deemed “preventable” by panels of clinicians advising the ACSQHC. The intent of the HAC developers and funding authorities is to “indicate” problems in patient safety outcomes by sampling those which by consensus should be preventable. Much like a medical biopsy, indicator sets are intended to diagnose problems from a small sample of the larger organ. These clearly identify and characterise diseased tissues in the sample, but are subject to sampling error if not adequately targeted.

By contrast, the CHADx+ is designed to be exhaustive of all hospital-acquired diagnoses, on the assumption that each clinical unit will select a relevant “reducible” subset of these for their ward or clinical specialty. This approach provides greater flexibility to support clinical quality improvement, by allowing clinical groups to focus their attention on directly relevant clinical measures rather than hospital-wide indicators. Comparative rates will allow health services to identify specialty units in other hospitals that are able to achieve better patient outcomes and learn from these units which clinical practices or pathways may lead to this improvement.

A comprehensive system like CHADx+ functions more like a positron-emission tomography (PET) scan than a biopsy. In the case of PET, the differential uptake of fluorodeoxyglucose (FDG) identifies neoplasms in a patient. The image produced is fuzzy, and may also reflect non-cancer processes, but it is able to detect secondary tumours in parts of the body that might not otherwise be biopsied. In the case of the CHADx systems, a data scan may identify some types of patient harm that are not modifiable with current medical knowledge or may be relatively insignificant from a clinical outcome perspective (e.g. nausea and vomiting). However, because it picks up the entire range of harm to patients in hospital, it facilitates variation analysis to identify those complications amenable to prevention.

The problems with using the concept of “preventability” are well known: modest inter-rater reliability (the “eye of the beholder” problem) (Hayward and Hofer, 2001; Localio et al., 1996), the fact that what is not preventable may change with new medical knowledge (Pronovost et al., 2006) and that most adverse outcomes of hospital care are reducible, even if not “preventable” in each instance. Prioritising complications that are always preventable has benefits in clinical governance, with the claim of preventability lending weight to incentives and penalties attached to these measures. However, limiting focus on this smaller set may not engage clinicians whose concern is with “my patients” in specialised areas of clinical practice.

Comprehensive systems also allow monitoring for the “squeezed balloon” problem, by enabling clinical teams to understand outcomes of interventions designed to improve patient care. Here the best example is the tension between deep vein thrombosis prevention (early ambulation) and falls prevention (reducing risks by delaying ambulation). Whichever intervention represents the “squeeze” on the balloon, clinicians would need to monitor closely for “ballooning out” of other adverse outcomes.

While coverage (targeted vs. comprehensive) is the biggest driver of differences between the two systems, specific decisions about code inclusions in the HAC definitions will result in differences between the two systems. HACs focus on serious injuries from falls; CHADx+ identifies all falls resulting in injuries. Excluding falls cases where clinical coders have omitted the Y92.22 “health service area” place of occurrence code (n = 277) will further reduce the HAC counts compared with CHADx+ which relies on the COF 1 to identify in-hospital cases.

The example of HAC 4.1 Post-operative haemorrhage/haematoma requiring transfusion and/or return to theatre shows the effect of code exclusion. By limiting the measure to the specific T81.0 bleeds “complicating a procedure” (further defined in notes as “haemorrhage at any site resulting from a procedure”), HACs will not report the 1015 instances in Victorian data that use the less specific symptom code R58 Haemorrhage not elsewhere classified, at least some of which may relate to post-operative haemorrhage, when documentation is lacking for the causal statement “resulting from…” In these circumstances, coders may choose the less specific code to avoid over-interpreting the record. If financial penalties are introduced (COAG, 2016), there is additional incentive for doctors and coders to use the less-specific R58 code to avoid such penalties.

Neither system can address the problem of acute exacerbation in hospital of a chronic condition. In theory, hospitals should be able to reduce the incidence of such events and a good classification would identify where they occur. While the Australian Coding Standard is clear with respect to those conditions which can be coded using a combination code, where at least one concept included in the code arises after admission, many chronic conditions do not have combination codes. The standard specifically instructs that previously existing conditions that are exacerbated during the current episode of admitted patient care are assigned a COF 2 Condition not noted as arising during the episode of admitted patient care (ACCD, ACS, 2015b; Standard 0048 Condition Onset Flag).

Exacerbations may be present on admission but undocumented as such, and it may be difficult to identify additional treatment necessitated by the exacerbation or to differentiate this treatment from that provided for the underlying condition, making it difficult even to develop a reliable flag for these circumstances. This problem also applies to the progression of conditions such as pressure injuries, where a patient admitted with a stage 1 pressure area may deteriorate over the course of the admission, but because the pressure injury was present on admission it may be assigned COF 2 Condition not noted as arising during the episode of admitted patient care.

Lower level hospital-acquired pressure injuries may be considered to be “near misses” for higher stages (3 and 4) injuries. As HACs report only these higher stages, the CHADx+ category for stages 1 and 2 may be useful for high dependency units and other wards as an overall measure of their pressure injury risks. If HACs are used as the basis for adjustments in funding models, attention should be paid to the clinical ambiguities in assigning stages 2 and 3, to avoid incentives to document these at the lower level. As noted for post-operative haemorrhage versus haemorrhage not elsewhere classified (NEC), coding vigilance should also apply to those HAC classes where the .9 extension (generally “unspecified” diagnoses) have been omitted from HAC code sets, as this may encourage less precise documentation and coding in the context of funding incentives.

Risk adjustment is frequently applied to quality and safety measures, especially when they are meant for publication. This is to prevent health services and clinicians from being unfairly criticised when they treat more seriously ill patients with resulting higher rates of complications. The HACs address this by limiting the focus of many classes to specific patient groups in both the numerator and denominator and by excluding some intrinsically high-risk patient groups (e.g. newborns <2000 g) altogether.

While the specifications for these adjustments are published, when they are applied for reporting, the risk adjustment factors may not be apparent. While the HAC risk adjustment algorithm does not have a large impact on numbers of complications identified, its systematic omission of mental health admissions may reflect or exacerbate the larger problem of stigmatisation of mental health patients and their worse clinical outcomes even for unrelated, non-mental health problems (McLeay et al., 2017).

CHADx+ is designed to use a risk stratification approach. Rather than a “black box” a priori risk adjustment, CHADx+ allows clinicians and other users to actively choose the factors that define higher risk patients (e.g. emergency admission) through a computerised data portal. Statewide or peer group rates are thus also standardised for these clinical characteristics to make like-for-like comparisons.

Summary and conclusion

Australian Health Ministers have agreed to adjust hospital funding on the basis of quality metrics.

The national Commission on Safety and Quality in Health Care has developed two systems for using coded hospital data to identify poorer patient outcomes (HACs and CHADx), with the latter further refined by the Victorian Department of Health and Human Services.

While a quarter of the codes used fully overlap in the two systems, 70% of the CHADx+ classes have no HAC equivalent.

Uptake of both systems will be improved by clarifying their purposes and potential usefulness.

A clearer delineation between the two Australian systems for safety and quality monitoring in hospitals will assist funders, clinicians, quality improvement professionals and health information managers to understand discrepancies in case identification between them. Both clinical governance and quality improvement are important functions in the healthcare system. The challenge, particularly in the face of funding incentives, will be to ensure the appropriate application of purpose-designed and complementary tools for these two tasks.

Footnotes

Authors’ note

This project was undertaken while the authors were employed by the Victorian Department of Health and Human Services and, later by the newly formed Victorian Agency for Health Information.

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors received no financial support for the research, authorship, and/or publication of this article.

References

Australian Commission on Safety and Quality in Health Care (ACSQHC) (2012) National Safety and Quality Health Service Standards (September 2012). Sydney: ACSQHC. Available at: https://www.safetyandquality.gov.au/wp-content/uploads/2011/09/NSQHS-Standards-Sept-2012.pdf (accessed 3 May 2017).

Australian Commission on Safety and Quality in Health Care (ACSQHC) (2017a) Governance. Available at: https://www.safetyandquality.gov.au/about-us/governance/ (accessed 3 May2017).

Australian Commission on Safety and Quality in Health Care (ACSQHC) (2017b) Classification of Hospital Acquired Diagnoses ICD-10-AM 10th revision. Sydney: ACSQHC, 2017. Available at: https://www.safetyandquality.gov.au/ourwork/information-strategy/indicators/classification-of-hospital-acquired-diagnoses/ (accessed 3 May 2017).

Australian Commission on Safety and Quality in Health Care (ACSQHC) (2016a) Hospital Acquired Complications: Proof of Concept Study. Available at: https://www.safetyandquality.gov.au/our-work/informationstrategy/indicators/hospital-acquired-complications/ (accessed 24 November 2016).

Australian Commission on Safety and Quality in Health Care (ACSQHC) (2016b) Hospital Acquired Complications specifications version 1.1 October 2016. Available at: https://www.safetyandquality.gov.au/hospital-acquired-complications-specifications-v1-1-oct-2016/ (accessed 24 November 2016).

Australian Consortium for Classification Development (ACCD) (2015a) The International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, Australian Modification (ICD-10-AM): Australian Classification of Health Interventions (ACHI), 9th ed. Sydney: Independent Hospital Pricing Authority.

Australian Consortium for Classification Development (ACCD) (2015b) The International Statistical Classification of Diseases and Related Health Problems, Tenth Revision, Australian Modification (ICD-10-AM): Australian Coding Standards, 9th ed. Sydney: Independent Hospital Pricing Authority.

Australian Institute of Health and Welfare (AIHW) (2017) METeOR Metadata Online Registry Episode of admitted patient care – condition onset flag, code N. Available at: http://meteor.aihw.gov.au/content/index.phtml/itemId/354816 (accessed 12 January 2017).

Australian Institute of Health and Welfare (AIHW) (2016) Admitted patient care 2014–15: Australian hospital statistics. Health services series no. 68. Cat. no. HSE 172. Canberra: AIHW.

10.

Council of Australian Governments (COAG) (2016) Meeting Communiqué 1 April 2016. Available at: https://www.coag.gov.au/sites/default/files/communique/COAG_Communique.pdf (accessed 3 May 2017).

11.

Cromarty

Parikh

Lim

. (2014) Effects of hospital acquired conditions on length of stay for diabetic patients. Internal Medicine Journal 44(11): 1112–1116.

12.

Duckett

Cuddihy

Newnham

. (2016) Targeting zero: supporting the Victorian hospital system to eliminate avoidable harm and strengthen quality of care (Report of the Review of Hospital Safety and Quality Assurance in Victoria). Available at: https://dhhs.vic.gov.au/file/1061/download?token=u0M0GD17 (accessed 2 May 17).

13.

Eagar

Sansoni

Loggie

. (2013) A Literature Review on Integrating Quality and Safety into Hospital Pricing Systems. Centre for Health Service Development, University of Wollongong and Australian Commission on Safety and Quality in Health Care. Available at: https://www.safetyandquality.gov.au/wp-content/uploads/2012/12/Literature-Review-on-Integrating-Quality-and-Safety-into-Hospital-Pricing-Systems1.pdf (accessed 2 May 17).

14.

Hayward

Hofer

(2001) Estimating hospital deaths due to medical errors: preventability is in the eye of the reviewer. JAMA 286(4): 415–420.

15.

Jackson

(2017) Final Report: Implementation of the CHADx+ Portal to Support Hospital Clinical Quality Improvement. Melbourne: Victorian Agency for Health Information.

16.

Jackson

Michel

Roberts

. (2009a) A classification of hospital acquired diagnoses for use with routine hospital data. Medical Journal of Australia 191(10): 544–548.

17.

Jackson

Michel

Roberts

. (2009b) Development of a validation algorithm for “condition onset” flagging. BMC Medical Informatics and Decision-Making 9(48). Available at: http://www.biomedcentral.com/1472-6947/9/48

18.

Jackson

Nghiem

Rowell

. (2011). Marginal costs of hospital acquired conditions: information for priority setting for patient safety programs and research. Journal of Health Services Research and Policy 16(3): 141–146.

19.

KPMG and the Australian Commission on Safety and Quality in Health Care (ACSQHC) (2013) Final Report: High Priority Complications. Available at: https://www.safetyandquality.gov.au/wp-content/uploads/2014/06/National-set-of-high-priority-hospital-complications-Dec-2013.pdf (accessed 24 November 2016).

20.

Localio

Weaver

Landis

. (1996). Identifying adverse events caused by medical care: degree of physician agreement in a retrospective chart review. Annals of Internal Medicine 125(6): 457–464.

21.

McLeay

Harvey

Romaniuk

MNM

. (2017). Physical comorbidities of post-traumatic stress disorder in Australian Vietnam War veterans. Medical Journal of Australia 206(6): j3.

22.

McNair

Borovnicar

Jackson

. (2009) Prospective payment to encourage system wide quality improvement. Medical Care 47(3): 272–278.

23.

Michel

Nghiem

Jackson

(2009) Using ICD-10-AM codes to characterise hospital-acquired complications. Health Information Management Journal 38(3): 18–25.

24.

Pearse

Matcevska

Jackson

(2013) Analysis of Hospital-Acquired Diagnoses and Their Effect on Case Complexity and Resource Use Health Policy Analysis Final Report. Sydney: ACSQHC.

25.

Pronovost

Needham

Berenholtz

. (2006) An intervention to decrease catheter-related bloodstream infections in the ICU. New England Journal of Medicine 355: 2725–2732.

26.

Trentino

Swain

Burrows

. (2013) Measuring the incidence of hospital acquired complications and their effect on length of stay using CHADx. Medical Journal of Australia 199(8): 543–547.