Defining Amenable Mortality / September 2016

Contents

Context

High level definition

Inclusion and exclusion criteria for amenable mortality...... 2

Amenable mortality code list (1 July 2016)...... 3

Mortality data...... 6

Population data

Standardisation of Amenable Mortality rates...... 7

Reporting

Interpreting amenable mortality rates

Contributory measures

Updating the amenable mortality indicator

Impact of the 2016 update of the Amenable Mortality code list

Commencement date

References..…………………………………………………………………………………………………… Acknowledgements……………………………………………………………………………………….11

Context

Amenable mortality was introduced as a system level measure for the New Zealand health systemon 1 July 2016.This metric is widely accepted internationally as a valid and reliable indicator of health system performance,and can serve to identify potential areas of concern for more detailed investigation (Ministry of Health 2016, Ministry of Health 2010).This report covers the definition and calculation of amenable mortality, and lists the current amenable cause of death codes.

High level definition

Amenable mortality is defined as premature deaths (deaths under age 75) that could potentially be avoided, given effective and timely healthcare. That is, early deaths from causes (diseases or injuries) for which effective health care interventions exist and are accessible to New Zealanders in need.

Not all deaths from these causes could be avoided in practice. For example, because of comorbidity, frailty and patient preference. However, a higher than expected rate of such deaths in a District Health Board (DHB) may indicate that improvements are needed with access to care, or quality of care.

Inclusionand exclusion criteria for amenable mortality

  1. A specific intervention, package of interventions, or model of care (hereafter ‘intervention’) must be identified and linked to a specific cause of death. There must be a clear International Classification of Disease(ICD) code for the cause of death. Reporting and coding of this cause of death must be of high quality.
  2. The intervention must be a medical or surgical intervention delivered by or under the direction of a clinician to patients or people at risk, in a healthcare setting or at home. The intervention may involve screening, diagnosis or rehabilitation, as well as treatment. Public health interventions delivered collectively to populations (e.g. food safety laws, tobacco taxes, safe sex social marketing campaigns) are excluded. This is so that the measure reflects access to and effectiveness of health care, rather than wider social systems.
  3. The intervention must have been introduced and become generally accessible to patients or at risk populations within the past 40–50 years (ie post 1960). This is because interventions introduced many decades ago are likely to have become diffused even in poorly performing health systems, so such interventions provide no comparative information regarding current health system performance.
  4. The intervention must have already reduced under 75 mortality (in the relevant NZ subpopulation or patient group) by >30%. Alternatively the intervention must have been shown in randomised controlled trials or high quality observational studies to be capable of such mortality reduction, within 5 years of universal coverage being achieved. The upper age limit reflects high prevalence of multimorbidity in the very old, making valid assignment of a single underlying cause of death difficult in this age group. A short lag period (<5 years) is required so that the metric indicates current, not future, health system performance. Note that some amenable cause of deaths may have age restrictions within the 0–74 age range.
  5. The linked cause of death must account (currently) for >0.1% of all under 75 deaths (approximately 10 deaths per year). This is an optional criterion and is not strictly enforced. It is included only to avoid cluttering the list with causes of death whose associated mortality has fallen to very low levels, with little possibility of resurgence absent total health system collapse – such causes of death are ‘avoided’ rather than ‘avoidable’.

Amenable mortalitycode list (1 July 2016)

The current list of amenable causes of death provides the causes of death which are understood to be avoidable through personal health care (table 1).This table was updated by an expert panel in June 2016. The updated list comprises 38 conditions, grouped into six super categories:

  • Infections
  • Maternal and infant conditions
  • Injuries
  • Cancers
  • Cardiovascular diseases and diabetes
  • Other chronic diseases.

Table 1: Current amenable mortality code list (July 2016)

Group / Condition / ICD-10-AM-VI / Notes
Infections / Pulmonary tuberculosis / A15-A16
Meningococcal disease / A39
Pneumococcal disease / A40.3, G00.1, J13
Hepatitis C (HCV) / B17.1, B18.2 / New
HIV/AIDS / B20-B24
Cancers / Stomach cancer / C16
Rectal cancer / C19-C21
Bone and cartilage cancer / C40-C41
Melanoma of skin / C43
Female breast cancer / C50
Cervical cancer / C53
Uterine cancer / C54, C55 / New
Prostate cancer / C61
Testis cancer / C62
Thyroid cancer / C73
Hodgkin lymphoma / C81
Acute lymphoblastic leukaemia / C91.0 / Ages 0-44
Maternal and infant disorders / Complications of pregnancy / O00-O96, O98-O99
Complications of perinatal period / P01-P03, P05-P94
Cardiac septal defect / Q21
Cardiovascular disorders and diabetes / Diabetes / E10-E14
Valvular heart disease / I01, I05-I09, I33-I37
Hypertensive diseases / I10-I13
Coronary heart disease / I20-I25
Pulmonary embolism / I26
Atrial fibrillation & flutter / I48 / New
Heart failure / I50
Cerebrovascular diseases / I60-I69
Other chronic disorders / Chronic obstructive pulmonary disease (COPD) / J40-J44
Asthma / J45-J46
Cholelithiasis / K80
Renal failure / N17-N19
Peptic ulcer disease / K25-K27
Injuries / Land transport accidents excluding trains / V00, V01-V04, V06-V14, V16-V24, V26-V34, V36-V44, V46-V54, V56-V64, V66-V74, V76-V79, V80.0-V80.5, V80.7-V80.9, V82-V86, V87.0-V87.5, V87.7-V87.9, V88.0-V88.5, V88.7-V88.9, V89, V98-V99
Accidental falls on same level / W00-W08, W18
Fire / X00-X09
Suicide / X60-X84

Defining Amenable Mortality1

Mortalitydata

The numerator data for the amenable mortality metric is extracted from the Ministry of Health’s Mortality Data Collection. The Mortality Data Collection uses information from a variety of sources.These include death certificates, hospital separations summaries, patient records, coronial reports and police reports.The data is used to code the underlying cause of death for every death in New Zealand.The underlying cause is defined as the cause that initiated the train of events leading to the death.

Currently, cause of death is coded using ICD-10 AM version VI. 2014 will be the first year coded in ICD-10-AM-VIII. Note that the Ministry changes versions regularly, and at some point ICD-10 will be replaced by ICD-11.

To calculate amenable mortality, all deaths registered in New Zealand in the relevant calendar year with an underlying cause of deathincluded in the current version of the amenable mortality codelist, where the deceased was aged 0–74 years at the date of death, are extracted.

Variables extracted from the Mortality Data Collection for each relevant death include:

  • National Health Index
  • DHB of residence (and linked NZDepdecile)
  • Date of birth (and age in five-year age bands)
  • Date of death
  • Sex
  • Ethnicity (prioritised)
  • Underlying cause of death (ICD-10-AM three or four digit code as applicable)

Population data

Denominator data to calculate amenable mortality rates are the national or DHB usually resident populations derived from projections the Ministry gets annually from Statistics New Zealand for the Population Based Funding Formula (PBFF).Statistics New Zealand populations are preferred to Primary Health Organisation enrolment data as the latter does not cover everyone usually resident in New Zealand. The ethnic denominators are based on prioritised ethnicity, to align with the numerator data. Both these denominators and the NZDepdecile or quintile denominators are derived internally within the Ministry from the Statistics New Zealand projections; they are not available directly from Statistics New Zealand.

Variables extracted include:

  • Year
  • DHB
  • Age 0–74 in five-year age bands
  • Sex
  • Ethnicity
  • NZDepdecile

Standardisation of amenable mortality rates

Crude amenable mortality rates (along with 95% or 99% confidence intervals) are calculated by dividing the amenable mortality count by the corresponding population count (with confidence intervals estimated in the usual way). However, these crude rates do not allow fair comparison of one DHB with another, or the same DHB with itself over time, because of variation in the underlying population age and sex structure, ethnic mix or socioeconomic (ie deprivation) distribution.

To control for confounding by these socio-demographic variables, a combination of stratification and standardisation is employed. Alternatively, regression modelling may be used. This is the ideal method but may not be practicable outside a research setting.

When comparing DHBs overall, the usual approach is to directly standardise for age. In this case, the reference population used as the source for the age weights shouldbe an ‘older’ population, as thiswill minimise distortion of DHB rates overall. A recent New Zealand population projection could be selected.Alternatively, the World Health Organization’s World Standard Population can be used as the reference for the age weights, if international comparison is needed.

Note that this controls for confounding by age, but not by other variables such as ethnic composition and socio-economic (ieNZDep) distribution. This would generally be done by indirect standardisation, which has the advantage of being robust with low counts and does not require a standard population. However, each DHB can then be compared only with the national population, not with one other (as each acts as its own ‘reference’). Alternatively, direct double standardisation can be used if the only major confounders are age and ethnicity; the same projected New Zealand population then acts as the reference or source for both age and ethnic weights (triple standardisation for age, ethnic mix and NZDep distribution is not recommended). However, the best way to control for confounding by multiple variables is regression modelling.

When examining indigenous or ethnic inequalities, the main population of interest is theMāori(or Pacific or Asian) population, so a population with a ‘young’ age structure should be used as the reference population for direct age standardisation (in order to avoid distorting the rates of interest). For this purpose, there is no need to adjust for socioeconomic variation (because deprivation is a mediator, not a confounder, of the ethnicity – mortality relationship), and the inequality estimates (age standardised rate ratios [SRR] or rate differences [SRD]) are already stratified by DHB. Variation in the ethnic SRR (or SRD) by DHB can then be directly assessed. For example if theMāori-nonMāoriamenable mortality SRR in Northland DHB is 2.0 and that in Auckland DHB is 1.5, then the indigenous inequality in amenable mortality in Northland is twice that in Auckland (or 100% greater):

[2.0 – 1.5] / [1.5 – 1]

While any population with a young age structure could serve as the standard, the 1996-2000 or the 2001 Māoricensalpopulation is often used. However, the Māori population (and Asian and Pacific populations) is undergoing rapid ageing and it makes little sense to use an obsolete population that no longer reflects the age structure of the population of interest. Rather, an appropriate reference population would be a recentMāori population projection (such as that derived by the Ministry for the PBFF). Alternatively, Segi's World Population could be used.This was the international standard for most of the last 50 years and is still used by the International Agency for Research on Cancer. While its age structure is slightly older than that of the currentMāori population, it may not be too far from that of theMāori population in (say) 2025 – which would provide a suitable standard for the next 20 years or so.

Note that whatever the source of weights, these should be the All Ages population (to allow wider comparisons) although weights normed to the 0–74 population are also acceptable.

Also note that any form of summarisation (whether through direct or indirect standardisation or any other modelling approach) may disguise important information. For example, ifMāori – nonMāori inequality in amenable mortality rates is different for children than for adults, this effect modification by age will be lost if only the age standardised rates are compared. Instead, fully stratified results (ie age/sex/ethnic/ NZDep specific rates) should be reported whenever there is adequate cell size to do so.

Reporting

Reports are made available to DHBs annually, in or around February each year, subject to mortality data being available in December. DHBs are provided with a rolling five year data set, covering the period from seven to two years prior to the current calendar year.

It takes several years for some coronial cases to return verdicts. Given the significant impact these cases can have on some amenable causes of death,estimates for this indicator are not available till approximately 2 years after the end of the year of death registration.

Interpreting amenable mortality rates

When interpreting amenable mortality rates there are some issues to be aware of:

  1. Differences between DHBs, or for the same DHB over time, may reflect residual confounding.Age standardisation or even double standardisation for age and ethnicity does not remove confounding by other variables.These could include deprivation distribution, rurality, or migration.
  2. The data sets are two to three years old, so may be out of date, and some areas of poor performance may already have been fixed.
  3. Differences may not be large enough to be clinically or epidemiologically important, even if statistically significant at the 99% level generally used for DHB level data.
  4. On the other hand, when disaggregated to specific condition level, numbers of deaths from some causes in some DHBs may be too small to permit any conclusions to be drawn.
  5. Even when not disaggregated by cause, the data may be too sparse to allow analysis at subdistrict level.Fundamentally, premature deathis a rare event.
  6. Similarly, some ethnic analyses may have insufficient volumes. Note that all subgroup analyses will be more subject to random variation from year to year than total DHB level analyses.
  7. Over time, medical advances lead to additional causes of death becoming classified as amenable (when the codelist is updated).Each update of the code listtherefore creates discontinuities in the time series, which can make interpretation of longterm trends in amenable mortality challenging.
  8. Trends and contrasts in amenable mortalitydo not necessarily imply that corrective action is required on the part of any DHB. However, a concerning trend or contrast is a warning signal. Such a finding should lead the relevant DHB to investigate further.The indicator could be signalling a failure or deterioration in performance of one or more of the DHB’s funded health services, or some other explanation may apply.
  9. It is recommended that DHBs look first at the amenable mortality metric per se, then drill down to super category level, and finally drill down to specific condition level. This should be done for the DHB population as a whole, and also separately for each lifecycle stage, sex, major ethnic group and NZDep quintile.
  10. If a particular cause of death, or group of cause of deaths, is identified as the cause of the concerning amenable mortality rate or trend, then further investigationof the service mainly responsible for managing patients with the relevant disease(s) may identify concrete corrective action the DHB or Alliance can undertake.
  11. In this case, part of the evaluation of implementation of the corrective action will be to monitor whether theunder-75 mortality rate from that condition, and the amenable mortality indicator as a whole, improves, allowing for the two year delay involved with reporting this indicator.

Contributory measures

A wide range of health services are involved in reducing amenable mortality rates. The major causes contributing to the amenable mortality rate vary by age group (within the 0–74 age range) and sex. Some of the key contributory measures that DHBs can focus on to improve their performance on this indicator may include:

  • Cancer screening coverage and treatment timeliness
  • Cardiovascular risk management
  • Other chronic disorder management (COPD, diabetes)
  • Injury prevention
  • Mental health services (self harm)
  • Smoking cessation services.

Updating the amenable mortality indicator

The amenable mortalitycode list requires regular updating because:

  • Medical advances mean that some causes of death not previously classified as amenable will now meet all inclusion criteria.
  • Mortality from some causes of death currently included as amenable will fall so low that no further fall is possible; nor would any recurrence in mortality from these conditions be expected absent total collapse of the health system. While leaving such causes of death on the list does no harm, it adds nothing either and does increase ‘clutter’ and opportunity for error.

Unfortunately, any change to the code list creates a discontinuity in the time series that makes interpreting trends in amenable mortality rates challenging. So the frequency of updating needs to balance these opposing tensions. For that reason, we propose to update the code list at 5–10 year intervals.

Impact of the 2016 update of the amenable mortalitycode list

The previous version of the code list was developed in 2009 (using data up to 2006). Accordingly, it was decided to update it in 2016 (using data up to 2013), given the inclusion of amenable mortality as a system level measure on 1 July 2016.

A clinical expert panel identified potential interventions introduced since 2006. The causes of death linked to these interventions were filtered through the inclusion and exclusion criteria listed above, and mortality trends were examined for the filtered causes. The expert panel then made final recommendations based on thisanalysis. The opportunity was also undertaken to assess whether any causes of death currently considered amenable should be dropped from the updated list. Relevant authorities in Australia (AIHW) and the UK (ONS) were also consulted to identify any new causes of death they were planning to include in their amenable mortalitycode lists.

This exercise resulted in the identification of three new causes of death and the decision to exclude one existing cause of death:

  • Inclusions: hepatitis C, atrial fibrillation & flutter, cancer of the uterus.
  • Exclusions: treatment injury (data quality inadequate)

Amenable mortality rates calculated using ‘old’ and ‘new’ lists showedthat the minor changes made to the code list in 2016 have not substantively affected the DHB rates, at least at whole-of-metric level. Nevertheless, it is important that the list be kept current, both for credibility and because differences may be larger at super category level, especially for some DHB population subgroups.