Report of the New South Wales Chief Health Officer	Table of contents On this page: Commentary References Print version
	Methods

• 1. Introduction
• 2. Data sets
  • 2.1 Health Outcomes Statistical Toolkit (HOIST)
  • 2.2 Population data
  • 2.3 Australian Bureau of Statistics Mortality Collection
  • 2.4 NSW Inpatient Statistics Collection
  • 2.5 NSW Midwives Data Collection (MDC)
  • 2.6 NSW Birth Defects Register
  • 2.7 NSW Health Surveys 1997 and 1998
  • 2.8 NSW Child Health Survey 2001
  • 2.9 NSW Population Health Survey
  • 2.10 ABS National Health Survey
  • 2.11 School surveys
  • 2.12 NSW Central Cancer Registry data
  • 2.13 NSW Notifiable Diseases database (NDD)
  • 2.14 Australian Childhood Immunisation Register (ACIR)
  • 2.15 Prison Inmate Survey
  • 2.16 Emergency Department Data Collection
• 3. Statistical Methods
  • 3.1 Crude death rates
  • 3.2 Age-adjusted rates
  • 3.3 Life expectancy at birth
  • 3.4 Life expectancy at age 65
  • 3.5 Analysis of NSW Health Surveys
  • 3.6 Graphical presentation
  • 3.7 Analysis of indicators by local government area
    • 3.7.1 Smoothing of estimates for population-based indicators
    • 3.7.2 Smoothing of estimates for binary-type indicators
    • 3.7.3 Interpretation of data in maps
• 4. Methods used for specific chapters and topics
  • 4.1 Health-related behaviours - Deaths and hospitalisations attributable to use of drugs and alcohol
  • 4.2 Rural and remote populations
  • 4.3 Country of birth
  • 4.4 Socioeconomic status
    • 4.4.1 Socioeconomic status measures
    • 4.4.2 Poisson regression models
  • 4.5 Avoidable mortality
  • 4.6 Avoidable hospitalisations
  • 4.7 Diabetes-related deaths
  • 4.8 Psychological distress
• 5. Area Health Service Boundaries
• 6. Quality assurance process
• 7. References
  
  
• Tables and figures
• Table 1. Australian standard population (30 June 2001)
• Table 2. K10 score and level of psychological distress
• Table 3. 1996 and 2005 Area Health Services
• Figure 1. Sample graph demonstrating point estimates and 95% confidence intervals.

1. Introduction

2. Data sets

2.1 Health Outcomes Information Statistical Toolkit

The Health Outcomes Information Statistical Toolkit (HOIST) is a SAS-based 'data warehouse' operated by the Centre for Epidemiology and Research of the NSW Department of Health. It brings together most of the data collections often used in population health surveillance in NSW, and contains all the available historical data for each collection. HOIST data are in one format - SAS datasets - and HOIST code values are, as far as possible, consistent across time and among datasets. HOIST provides a common data analysis environment across the public health network in NSW.

2.2 Population data

Population estimates as at 30 June were used for calendar years, while estimates as at 30 December were used for financial years. Age- and sex-specific estimated resident populations (ERPs) for NSW Statistical Local Areas (SLAs) at 30 June were obtained from the Australian Bureau of Statistics (ABS) for use with calendar year data. A cubic spline interpolation between mid-year ERPs was used to derive 30 December age- and sex-specific population estimates for use with financial year data. Populations of NSW area health services were derived by aggregating the appropriate SLA-level ERPs, except in the case of Sydney South West and South Eastern & Illawarra Area Health Services, the border between which transects 2 SLAs. ERPs for these SLAs were apportioned according to the proportions derived from the usual resident counts from the 2001 Census at the collection district level.

The 2001 Australian mid-year ERP, shown in Table 1, was used as the standard population for age-adjustment.

Table 1. Australian standard population (30 June 2001)
Age	Persons	Age	Persons
0-4 yrs	1,282,357	50-54 yrs	1,300,777
5-9 yrs	1,351,664	55-59 yrs	1,008,799
10-14 yrs	1,353,177	60-64 yrs	822,024
15-19 yrs	1,352,745	65-69 yrs	682,513
20-24 yrs	1,302,412	70-74 yrs	638,380
25-29 yrs	1,407,081	75-79 yrs	519,356
30-34 yrs	1,466,615	80-84 yrs	330,050
35-39 yrs	1,492,204	85+ yrs	265,235
40-44 yrs	1,479,257
45-49 yrs	1,358,594	All ages	19,413,240
Source: ABS population estimates (HOIST), Centre for Epidemiology and Research, NSW Department of Health.

Country-of-birth (COB)-specific populations used for country of birth pages were derived from annual age-, sex- and COB-specific ERPs for all of Australia supplied by the Australian Bureau of Statistics. Equivalent populations for NSW were derived from these national populations by estimating the proportion of immigrants in each age, sex, and COB stratum who reside in NSW, based on cubic splines fitted to age-, sex-, COB- and specific State-Territory counts from the 1981, 1986, 1991, 1996 and 2001 ABS Censuses of Population and Housing, and then applying these proportions to the national age-, sex-, and COB-specific ERPs for each year. This was done to ensure that the COB-specific populations for NSW were based on estimated residential populations; and that the non-linear, and often dramatic, changes in immigrant populations in inter-Censal periods were accurately reflected in the NSW population estimates used.

2.3 Australian Bureau of Statistics Mortality Collection

For this report, ABS mortality data for deaths of NSW residents registered anywhere in Australia were accessed via HOIST. Deaths are presented by calendar year of death.

All deaths for which a coronial inquiry is not required must be certified by a registered medical practitioner as to cause and date; the certificate is registered by the registrar of births, deaths and marriages in each state or territory. Most deaths due to accidental causes, deaths occurring under suspicious circumstances (in which foul play cannot be excluded), deaths occurring shortly after anaesthesia or surgery, and deaths of persons who had not been seen by a medical practitioner in the year preceding their death, automatically become coronial cases and are registered by a coroner at the conclusion of an inquiry into the circumstances of the death.

Most non-coronial deaths are registered with the relevant registrar of births, deaths and marriages within 4 weeks of the date of death. However, coronial inquiries can take months, and in some cases years, to conclude. Mortality data are supplied by the ABS by year of registration. Therefore, deaths occurring in the last few weeks of each calendar year (or the last few months for coronial cases) may not be registered until January in the subsequent year. Delays in registering deaths tend to be greater for some causes of death, and for people resident in rural areas.

At the time of preparation of this report, the most recent mortality data available from ABS included only those deaths registered in 2004. The 2004 numbers were adjusted to include an estimate of the number of deaths due to that cause that occurred in 2004 but were not registered until 2005. A pro rata adjustment was made, based on registrations for the preceding three years (2001 to 2003). The first step was to determine the proportion of total deaths in the preceding three years which were not registered until the following year. That proportion was used to multiplying the number of deaths registered in 2002, to obtain the estimate of the number of deaths still to be registered. The estimates were calculated for each age-sex stratum. Where deaths were further categorised, for example by geographical place of residence or country of birth, the imputation procedure was carried out separately for each category.

For deaths registered during or before 1996, a single code for the principal underlying cause of death (based on the information recorded on the death certificate by a medical practitioner or coroner) was selected for each death. For deaths registered since 1997, the ABS has used computer-assisted cause-of-death coding that yields up to 20 contributing causes of death in addition to the principal underlying cause of death.

From 1999 onwards, causes of death have been classified according to the 10th revision of the International Classification of Diseases (ICD-10, World Health Organization, 1992). Deaths registered before 1999 were coded according to the 9th revision of the International Classification of Diseases (ICD-9, World Health Organization, 1977).

The ICD-10 and ICD-9 codes used for each indicator are included in disease and procedure codes appendix.

2.4 NSW Inpatient Statistics Collection

The NSW Inpatient Statistics Collection (ISC) or Admitted Patient Data Collection is a census of all services for admitted patients provided by public hospitals, public psychiatric hospitals, public multi-purpose services, private hospitals and private day procedure centres in NSW. The ISC is a financial year collection from 1 July through to 30 June of the following year. The information it contains is provided by patients, health service providers, and the hospital's administration. The information reported includes patient demographics, source of referral to the service, service referred to on separation, diagnoses, procedures, and external causes.

The ISC includes data on hospital admissions of NSW residents which occurred in hospitals interstate. The only exception to this is that data from interstate hospitals for the year 2004-2005 are not yet available. This may affect analyses involving uncommon diagnoses or procedures, particularly in Health Area analyses, and has a greater effect on rates for areas closer to an interstate boundary. The number of interstate admissions has been estimated for 2004-05, based on admissions for the preceding five years (1999-00 to 2003-2004). The first step was to determine the proportion of total admissions for NSW residents in the preceding five years which were at interstate hospitals. That proportion was used to multiply the number of admissions at hospitals in NSW, to obtain the estimate of the number of admissions expected to have occurred at interstate hospitals. The estimates were calculated for each age-sex stratum. Where hospitalisations were further categorised, for example by geographical place of residence or country of birth, this imputation procedure was carried out separately for each category, thus accounting for the uneven distribution of interstate hospital admissions.

For this report, the ISC was accessed via HOIST. From 1 July 1998, ISC data on HOIST have been for episodes of care in hospital. Episodes of care end with the discharge, transfer, or death of a patient. A new episode of care may also start when the service category for an admitted patient is altered, as a result of a change in the on-going clinical care requirements for that patient during the one episode of accommodation in a single facility. ISC data on HOIST up to 30 June 1998 were for periods of stay in hospital. A period of stay in hospital ends with the discharge, transfer, or death of a patient, and may consist of multiple episodes of care. The change from "period of stay" to "episode of care" causes a small increase in the apparent number of admissions.

From 1 July 1998, the reason for a hospital admission has been coded at the time of separation (discharge, transfer or death), according to the 10th revision of the International Classification of Diseases, Australian Modification ICD-10-AM (National Centre for Classification in Health, 2000). Prior to this, it was coded according to the 9th revision of the International Classification of Diseases, Clinical Modification (ICD-9-CM), using the Australian version (National Coding Centre, 1996) from July 1995 and the US version prior to that.

From 1 July 1998, procedures carried out during a patient's stay have been coded according to the MBS-Extended Procedure Classification, published as Volume 3 and Volume 4 of the 10th revision of the International Classification of Diseases, Australian Modification ICD-10-AM (National Centre for Classification in Health, 2000). Prior to this, procedures were coded according to the 9th revision of the International Classification of Diseases, Clinical Modification (ICD-9-CM), using the Australian version (National Coding Centre, 1996) from July 1995 and the US version prior to that.

The numbers of diagnosis and procedure codes that may be recorded, at the time of separation, have varied over time, and are currently as follows:

· principal diagnosis (the principal reason for admission);

· up to 55 other diagnoses;

· up to 50 procedures and procedure blocks;

· up to eight external cause codes for injury and poisoning.

· up to three codes for place of occurrence injury or poisoning.

· up to three codes for activity at time of injury or poisoning.

Extensive use of mapping tables between ICD-9-CM and ICD-10-AM disease codes, produced by the National Centre for Classification in Health, was made to obtain the most appropriate match for individual codes between the 2 classification systems. The ICD-10-AM and ICD-9-CM codes used for each indicator are included in the disease and procedure codes appendix.

2.5 NSW Midwives Data Collection

The New South Wales Midwives Data Collection (MDC) is a population-based collection covering all births in NSW public and private hospitals, as well as home births. It does not receive notifications of interstate births where the mother is resident in NSW.

The data collection has operated continuously since 1990. It encompasses all livebirths and stillbirths of at least 20 weeks gestation or at least 400 grams birthweight. The MDC relies on the attending midwife to complete a notification form when a birth occurs. The form includes demographic items, and items on maternal health, the pregnancy, labour, delivery, and perinatal outcomes. It has undergone 3 revisions over the years. The Midwives Data Collection database is compiled in the Information Management and Support Branch of the NSW Department of Health.

For this report, the MDC was accessed via HOIST. Data are presented for calendar years. Data for mothers who gave birth in NSW but reside in other states was excluded. The MDC does not include data for mothers who live in NSW but give birth outside the state.

The key indicator of perinatal deaths in the report uses data derived from the Australian Bureau of Statistics. Perinatal deaths among infants of at least 22 weeks gestation or 500 grams birthweight are reviewed by the NSW Maternal and Perinatal Committee. Both stillbirths and neonatal deaths are classified according to an obstetric cause-specific classification, Perinatal Society of Australia and New Zealand Perinatal Death Classification (PSANZ-PDC). Neonatal deaths are also classified by neonatal cause according to the Perinatal Society of Australia and New Zealand Neonatal Death Classification (PSANZ-NDC).

2.6 NSW Birth Defects Register

The NSW Birth Defects Register (BDR) was established in 1990. Legislation to mandate the notification of birth defects recognised at up to 1 year of age, has been in effect since 1 January 1998. Prior to that, the BDR operated on a voluntary reporting basis.

For this report, the BDR was accessed via HOIST. Data are presented for calendar years.

2.7 NSW Health Surveys 1997 and 1998

In 1997 and 1998, the NSW Department of Health, in conjunction with the 17 area health services at that time, conducted 2 population health surveys using computer-assisted telephone interviews (CATI) (Williamson et al., 2001). The main aims of the surveys were to provide local and statewide information to inform health service planning and policy development. The survey questions focused on the 6 NSW health priority areas: cardiovascular disease, cancer, mental health, injury, diabetes, and asthma.

The target sample for each year comprised 1000 NSW residents aged 16 years and over from each of the 17 NSW Health Areas at that time (total sample 17,000 people each year). A stratified two-stage cluster sample design was used, with simple random sampling of all potentially active telephone numbers within each NSW health area, and simple random sampling of one household resident for interview.

Interviews were conducted in 6 languages (English, Arabic, Chinese, Greek, Italian, and Vietnamese) by trained interviewers at the NSW Department of Health's CATI facility.

The total sample size was 35,027 respondents (17,531 in 1997; 17,496 in 1998). The overall response rate for both surveys was 70%.

For this report, data from the NSW Health Surveys 1997 and 1998 were accessed via HOIST.

2.8 NSW Child Health Survey 2001

In 2001, the NSW Department of Health, in conjunction with the 17 area health services at that time, conducted a survey of the health of children in NSW, using CATI.

Development of the survey instrument was overseen by a technical reference group. The final questionnaire covered topics including use of health services, nutrition, food security, asthma, oral health, parent support services, social support, sun protection, sight, hearing, speech, family functioning, social capital, smoking, sports and other organised activities, and physical activity.

The target sample comprised at least 500 NSW children aged 0-12 years from each of the then 17 NSW health areas. Households were sampled using a method similar to the 1998 NSW Health Survey (Williamson et al., 2001). One eligible child was selected from each household, using random numbers generated by the CATI system. A parent or carer of the selected child was interviewed.

Interviews were conducted in 4 languages (English, Arabic, Chinese and Vietnamese) by trained interviewers at the NSW Department of Health's CATI facility. A total of 9933 interviews were completed, while 1770 households or selected respondents refused to participate. This yielded a response rate of 84.9%.

For this report, data from the NSW Child Health Survey 2001 were accessed via HOIST.

2.9 NSW Population Health Survey

From 2002, the NSW Department of Health, in conjunction with the area health services, conducted the NSW Population Health Survey, an ongoing survey of the health of people in NSW using computer-assisted telephone interviewing (CATI). The main aims of the New South Wales Population Health Survey are to provide detailed information on the health of the people of NSW, and to support the planning,implementation, and evaluation of health services and programs in NSW.

The target population for the NSW Population Health Survey is all NSW residents living in households with private telephones. The target sample comprised approximately 1,200 people in each area health service (total sample of 12,000). When households were contacted, one person was selected, using random numbers generated by the CATI system.

Interviews were carried out continuously between February and December 2002 each year. Trained interviewers at the NSW Health Survey facility carried out interviews. When a child under the age of 16 years was selected, the main carer, known as the ‘proxy respondent’, was interviewed on behalf of the child. Most respondents were interviewed in English. The remaining interviews were conducted in Arabic, Chinese, Greek, Italian, and Vietnamese.

In 2005, 15,442 interviews were conducted, with 12,622 with people aged 16 years or over. The overall response rate was 67.6 per cent (completed interviews divided by completed interviews and refusals).

For this report, data from the NSW Health Survey were accessed via HOIST.

2.10 ABS National Health Survey

The National Health Surveys (NHS) conducted by the ABS collect information on illness and injury, health care use, and health risk factors. Data from both the 1989-90, 1995 and 2001 National Health Surveys are presented in this report. Data were accessed via HOIST, or were obtained as special tabulations from the ABS, or were from published reports.

2.11 School surveys

The NSW Department of Health and The Cancer Council NSW has carried out surveys of the health of secondary school students since 1984 as part of the triennial Australian School Students’ Alcohol and Drug (ASSAD) survey.

In its earlier years, the ASSAD Survey questions targeted drug and alcohol use. The topics covered by the Survey have gradually extended to include other issues that are important to the health of adolescents. In 2005 the NSW School Student Health Behaviours Survey included questions on physical activity and injuries, sun protection behaviours, eating behaviours, and mental health and wellbeing, in addition to smoking, alcohol and other drug use.

The information presented in this report may differ from information presented in previous Reports of the Chief Health Officer, which used a variety of published reports to collate information on trends. These published reports used data obtained from different surveys, which were then analysed using a variety of methods. For this report all recent and historical ASSAD/NSW School Student Health Behaviours Survey data were available for analysis for the first time. ASSAD/NSW School Student Health Behaviours Survey data has been collected in a consistent way over time and is the most reliable current source of information on trends in secondary school students’ health.

2.12 NSW Central Cancer Registry data

The NSW Central Cancer Registry was established by the NSW Department of Health in 1971 under the NSW Public Health Act. It was administered by the NSW Cancer Council, under contract, from 1986 until June 2004. The Registry has been managed by the Cancer Institute NSW since June 2004.

Notification of all newly-diagnosed cases of, and deaths due to, malignant neoplasm by hospitals and the Registrar of Births, Deaths and Marriages has been compulsory since the registry began. In 1991 the Act was amended to make notification by pathology laboratories compulsory as well. Notification has traditionally been via a printed notification form, although in recent years electronic notification by hospitals (but not pathology laboratories) has been introduced.

A case of cancer is the occurrence of a malignant neoplasm in one organ of a particular person. Therefore, a case of malignant melanoma in a particular person counts as one case. If the same person subsequently develops leukaemia, the leukaemia counts as a second case.

Incident cases and deaths registered before July 1999 were classified according to the 9th revision of the International Classification of Diseases (ICD-9, World Health Organization, 1977). Cases registered from July 1999 onwards have been classified according to the 2nd edition of the International Classification of Diseases for Oncology (ICD-O-2, World Health Organization, 1990). ICD-O-2 codes were translated back to ICD-10 codes by the Registry and used in this report.

For this report, cancer incidence data were accessed via HOIST. The cancer mortality data presented come from ABS mortality data.

2.13 NSW Notifiable Diseases Database

The NSW Notifiable diseases database (NDD), formerly called the NSW Infectious Diseases Surveillance system (IDSS), is a networked database used by 17 public health units (PHUs) located across NSW to register communicable disease notifications. Under authority of the NSW Public Health Act 1991, the NSW Health Department receives notifications of communicable disease via PHUs from general practitioners, hospitals, and pathology laboratories. Data are transferred weekly from PHUs to the Department, for compilation of statewide data. The Department, in turn, transfers a limited dataset to the Communicable Diseases Network of Australia and New Zealand (maintained by the Commonwealth Department of Health and Aging).

For this report, the NDD collection was accessed via HOIST.

2.14 Australian Childhood Immunisation Register

The Australian Childhood Immunisation Register (ACIR), which is managed by the Health Insurance Commission (HIC) and commenced operation on 1 January 1996, is a register of the immunisation status of all children less than 7 years of age. A Commonwealth-State cost-shared payment is made to service providers for data. Broadly, the functions of the ACIR are: to collect immunisation information from immunisation providers, and to administer a payments system to providers for reporting information; to provide immunisation status information to parents and providers and to administer a national recall-reminder service to parents; and to provide immunisation coverage data.

ACIR supplies NSW Health with monthly coverage data that identifies children 'overdue' for immunisation, which are forwarded to Public Health Units for follow up, and quarterly coverage data by local government area. These latter data form the basis for the information presented in this report.

2.15 Prisoner health survey 2001

The overall aims of the study were to describe the health of adult inmates, to identify factors associated with poor health, to develop indicators allowing comparisons to be made with the health of the general population, and to develop health goals and targets for the inmate population based on the findings. Special emphasis was placed on determining the health of Aboriginal and elderly inmates.

The methodology for the 2001 survey was similar to that used in 1996 to ensure consistency across the surveys. The design represented a cross-sectional random sample of inmates stratified by sex, age and Aboriginality. The sample included approximately 10% of male and 34% of female inmates in full-time custody. The survey was conducted between July and November 2001. All 29 correctional facilities in NSW were included in the survey.

According to the 2001 Inmate Census, there were 514 female and 7160 male prisoners in full-time custody on the 30th June 2001. Aboriginal people are over represented in the correctional system, comprising 16% and 25% of male and female prisoners compared with approximately 1% in the general community. Given this overrepresentation, and variations in health status between Indigenous and non-Indigenous Australians, it was decided to stratify by Aboriginality. The sample was also stratified into three age groups: under 25 years, 25 – 40 years, and over 40 years. The stratification ensured that there were sufficient numbers of both Indigenous and non-Indigenous inmates to enable the health status of each sub-group could be described separately. The overall response rate was 85% (700 males and 154 females). The response rates were 84% for women and 85% for men; and 83% for Aboriginal people compared with 85% for non-Aboriginal people.

2.16 Emergency Department Data Collection

The NSW Emergency Department Data Collection is a database of information collected from approximately one-third of NSW Emergency Departments. It represents approximately two-thirds of all NSW emergency patients. Analyses presented in this report are based on a provisional diagnosis assigned by staff when a patient presented to the Emergency Department.

For this report, data from the Emergency Department Data Collection were accessed via HOIST.

3. Statistical methods

3.1 Crude death rates

The crude death rate is an estimate of the proportion of a population that dies in a specified period. It is calculated by dividing the number of deaths in a specified period by the number at risk during that period (typically per year). It does not take into account the age structure of the population studied, and can be misleading when long-term trends are examined - or geographic areas are compared - because age structures of populations may vary over time or among areas. Crude death rates presented in this report used ABS estimated resident populations (ERPs) as at 30 June each year, and are expressed per 100,000 population per year.

3.2 Age-adjusted rates

Age-adjustment of rates used direct age-standardisation. This method adjusts for effects of differences in the age composition of populations across time or geographic regions. The directly age-standardised rate is the weighted sum of age-specific (five-year age group) rates, where the weighting factor is the corresponding age-specific standard population. For this report, the Australian estimated residential population (persons) as at 30 June 2001 was used as the standard population (this is given in Table 1). The same population was used for males and females to allow valid comparison of age-standardised rates between the sexes.

Ninety-five per cent confidence limits around the directly standardised rates were calculated using the method described by Dobson et al. (1991). This method gives more accurate confidence limits than the usual normal approximation for rarer conditions. Where the number of events is larger, the limits are equivalent to those calculated in the conventional fashion (Armitage, Berry and Matthews, 2002).

3.3 Life expectancy at birth

Life expectancy at birth is an estimate of the average length of time (in years) a person can expect to live, assuming that the currently prevailing rates of death for each age group will remain the same for the lifetime of that person. In fact, death rates will almost certainly change over the lifetime of a person born now, owing to changes in social and economic conditions, changes in lifestyle, advances in health care, and possibly the emergence of new diseases. However, because no-one knows what the death rates for each age group and sex will be in the future, the usual practice is to use the current rates of death to calculate life expectancy.

For this report, estimates and confidence intervals for life expectancy were calculated using abridged current life tables based on five-year age groups, except for the first 5 years of life, which were split into 2 age groups 0-<1 years and 1-4 years. The methods used are described in detail by Chiang (1984).

3.4 Life expectancy at age 65

The average number of additional years a person who has reached the age of 65 would expect to live if current mortality trends continue to apply is based on the age-specific death rates for a given year. This measure assumes that death rates will remain constant for the next 20 to 30 years, a much more conservative assumption than the one used to calculate life expectancy at birth. For this report, life expectancy was calculated using abridged current life tables based on 5-year age groups.

3.5 Analysis of NSW Health Surveys 1997 and 1998, NSW Child Health Survey 2001 and NSW Population Health Survey data

The survey samples were weighted to adjust for differences in the probabilities of selection among respondents, according to the number of eligible respondents in the household, and the number of residential telephone connections for the household (except in 1997, where telephone connection information was not collected). Post-stratification weights were used to adjust for differences between the age and sex structure of the survey samples and the relevant ABS mid-year population estimates (adjusted to exclude people resident in institutions) (Williamson et al., 2001).

The 'Surveymeans' procedure in SAS for Windows Version 8.02 was used to calculate point estimates and 95% confidence intervals. This procedure uses the Taylor expansion method to estimate sampling errors of estimators based on a stratified random sample (SAS, 2001).

3.6 Graphical presentation

Figure 1 below demonstrates the method used for graphical presentation of point estimates with their 95% confidence intervals. It shows age-adjusted incidence for melanoma for the years 1998 to 2002 for each of the NSW health areas. The standardised rate for NSW as a whole is indicated by the vertical reference line. The standardised rate, with its 95% confidence limits, for each health area, is shown as a horizontal line, with a central box indicating the point estimate.

Figure 1. Sample graph demonstrating point estimates and 95% confidence intervals.

3.7 Analysis of indicators by local government area

There are 17 LGAs with total populations less than 3000, and of these five have populations less than 2000 (based on population estimates as at June 2005). As the standard errors of estimates of rates or ratios are inversely related to population size this means that the standard errors vary greatly in their size between different LGAs. It also means that these estimates will vary greatly in the effect that chance events will have on the rate or ratio. Those with the smallest populations are particularly vulnerable to this. An extra one or two cases in a population of 2000 gives a far greater effect than the same increase in cases in a population of 200,000.

There also may be periods of time when there are no events within particular small areas, which would result in a rate for that area of zero. Two alternatives are possible here: to increase the number of years included in the analysis, preferably to 5 or more to obtain a reasonable estimate, or to use reasonably short periods of time (2 or 3 years) but borrow strength from adjacent areas and the variability across the entire state by using Bayesian methods. It is this latter method of "statistical smoothing" that we have applied in this report. Statistical smoothing allows calculation of an estimate for all areas, even when there are no cases/events in a particular area during the observed period of time; takes into account variability among areas (variously called global, uncorrelated or non-spatial variability), and more local effects (so-called spatial or local variability) in creating the estimate for individual areas; and 'borrows strength' from other areas, to improve the estimate in any particular area. This is more obvious in those areas with small populations.

There are two distinct methods of Bayesian smoothing used in this report, depending upon whether the indicator involves a population-based denominator (such as most indicators involving hospital separations or deaths)or whether it is of a binary nature (for instance smoking/not smoking in pregnancy). Estimation of population-based indicators are based on the indirectly standardised ratio obtained from indirect age and sex standardisation. In comparison, this report presents directly standardised rates for these indicators when considered on a statewide or health-area basis. Indirect statdardisation is used firstly, because age-specific rates required by direct standardisation can be unreliable in small areas where there are small numbers of cases; and secondly, the standardised mortality, or standardised incidence ratio (SMR or SIR) is conventionally modelled using Bayesian methods. The indirectly standardised ratio compares the number of cases in the local area with the number expected given the age-specific rates across the state and the age-specific population of the small area of interest. It is effectively an estimate of the relative risk of being in the disease group within that area compared to the state average risk, which by definition is unity (1). Indirect standardisation assumes that the pattern of age-specific rates across the entire population is appropriate for all small areas, which, although not tested, is usually a safe assumption.

3.7.1 Smoothing of estimates for population-based indicators

The smoothing used for the population-based indicators is obtained by applying the convolution or Besag, York and Mollie (BYM) model (Lawson et al, 2003). This model incorporates both spatially correlated and uncorrelated variation. It accounts for variability across the entire state (uncorrelated variation) as well as variability amongst the local government areas immediately adjacent to the area in question (spatially correlated variation). It is a fully Bayesian model which has been used substantially for disease mapping since it was introduced by Besag, York and Mollie (1991). Under the BYM model, the smoothed SIR/SMR (or relative risk) is implemented using Gibbs sampling within WinBUGS. The sample values for the parameter of interest (q_i) obtained by running this model in WinBUGS range in value within each area. The values form a distribution, which is known as the posterior distribution of q_i, and shows the expected distribution of the SIR/SMR for each area when adjusted (smoothed) for the two types of variability mentioned above. For each area the mean of this posterior distribution was used as the best estimate of the smoothed SIR/SMR, and the proportion of the probability distribution above unity was used as an estimate of the significance of the small area estimate relative to the state average. It is to be noted that the posterior distribution is dependent upon the expected number of cases: the higher the expected number of cases the smaller will be the standard error of the distribution, and hence the distribution will be 'tighter' around its mean.

A Bayesian 95% credible interval is analogous to the more common 95% confidence interval used in frequentist-based analyses. The 95% credible interval is obtained by determining the 2.5^th and 97.5^th percentiles of the posterior distribution. It can be interpreted as the range in which 95% of the estimates are located. The Bayesian method of obtaining an estimate of significance was also used by determining the proportion of the sampled values in the posterior distribution that lie above (or below) unity. A two-sided alternate hypothesis was used, so that if the proportion of the posterior distribution below unity was less than 0.025, then that area was considered to have significantly increased risk at the 5% level of significance.

The tabular output has been grouped by health area. Sydney LGA has been allocated to both Sydney South West and South East Sydney and Illawarra AHS, as this LGA is split between the two health areas. The columns in the output present the smoothed number of hospitalisations or deaths per year; the stabilized estimate of the Standardised Separation or Mortality Ratio, obtained by using Bayesian smoothing; the upper and lower 95% credible interval endpoints for the smoothed estimates of SSR/SMR; and the level of significance in relation to the state average indicated as follows:

++	means more than 99% of the distribution is above one, which would be equivalent to being significantly higher than the state average at 1% level of significance.
+	means more than 95%, but less than 99% of the distribution is above one, which would be equivalent to being significant at 5%.
0	means that between 5 and 95% of the distribution is above one, indicating that the SMR/SIR for this area is not significantly different to the state average of unity.
-	means less than 5% of the distribution is above one, which is equivalent to saying that the distribution is significantly lower than the state average at 5% level of significance.
--	means less than 1% of the distribution is above 1, which is equivalent to saying that the distribution is significantly lower than the state average at the 1% level of significance.

3.7.2 Smoothing of estimates for binary-type indicators

These indicators do not use a population-based denominator. Usually the cases or events of interest are a subset of all cases, so the denominator is obtained from the same source as the cases of interest. Most of these indicators are binary in nature, for example smoking in pregnancy, where the mother either answers yes or no to the question ‘did you smoke at all during pregnancy?’, or else are made into a binary variable, for instance whether antenatal care was commenced before 20 weeks or not. Because of the binary nature of these indicators, the most appropriate model is the logistic model.

The smoothing used for these indicators is obtained by modelling the data using a binomial distribution with a logit link function (Lawson et al, 2003). The model still incorporates both spatially correlated and uncorrelated variation, however we have not included age-standardisation. Implementation occurred using Gibbs sampling within WinBUGS. The sample values for the parameter of interest (smoothed proportion) range in value within each area. The values form a distribution, which is known as the posterior distribution of the smoothed proportion. For each area the mean of this posterior distribution is used as the best estimate of the smoothed proportion.

In order to compare areas and make mapping consistent with the other indicators, a posterior distribution for the prevalence ratio was calculated for each area by dividing each sampled value of the smoothed proportion by the overall proportion for the state. The proportion of the posterior distribution of this prevalence ratio above unity was used as an estimate of the significance of the small area estimate relative to the state average. The posterior distributions are dependent upon the total number of cases: the higher the total number of cases the smaller will be the standard error of the distribution, and hence the distribution will be 'tighter' around its mean.

Ninety-five percent credible intervals for the smoothed estimates of the proportion and the prevalence ratio as well as the estimate of significance were obtained using similar methods to the population-based indicators in 3.7.2 above. The tabular output included the raw number of cases (the event of interest) (if this was less than five, it was suppressed); the smoothed percent; the smoothed estimate of the prevalence ratio (based on distribution of the ratio of smoothed proportion to the state proportion, as given after Bayesian smoothing, but is not adjusted for the different age structures in the areas); the lower and upper 95% credible interval for ratio; and the level of significance (using the same symbols as for 3.7.2).

3.7.3 Interpretation of data in maps

The smoothed standardised incidence/mortality ratio is mapped for all indicators. The intensity of the colour of an area increases as the SIR/SMR increases, and the same scale is used for all maps. The maps show which areas are significantly different from the state average (at the 5% level of significance) by using ‘-‘ or ‘+’ to denote this. If an area does not differ significantly from the state, no symbol is shown. Maps were produced using SAS^® V9.1.3.

4. Methods used for specific chapters and topics

4.1 Health-related behaviours-Deaths and hospitalisations attributable to use of drugs and alcohol

Estimates of the numbers and rates of deaths and hospitalisations attributable to the use of tobacco, alcohol, and illicit drugs used aetiologic fractions developed by Ridolfo and Stevenson (2001). These fractions represent a revision of those originally published by Holman (1990) and later revised by English et al. (1995). They were derived from meta-analysis of published scientific literature on the adverse health effects (and in a small number of instances, protective effects) of these substances to estimate the proportions of cases of specific diseases and injuries that could be attributed to each substance.

For this report, an electronic file of the aetiologic fractions developed by Ridolfo and Stevenson was obtained from the Australian Institute of Health and Welfare. The disease and injury groupings used in this file were defined using ICD-9 and ICD-9-CM. The appropriate groupings of ICD-10 and ICD-10-AM codes were developed for this report. The codes used can be found in the disease and procedure codes section of the electronic version of this report.

Aetiologic fractions were applied to ABS Mortality data for NSW for the period 1989-2002, and NSW Inpatient Statistics data for the period 1989-90 to 2002-03.

4.2 Rural and remote populations

The chapter on rural and remote populations presents a range of health indicators for NSW according to ARIA+, the new enhanced Accessibility-Remoteness Index of Australia classification. This classification differs from the ARIA classification used in the previous report in several ways. In the ARIA classification an ARIA category was allocated on the basis of an average index score from 0-12 within each statistical local area (SLA). This index score was based on the road distance from the closest service centre in each of 4 classes. Under ARIA the smallest service centre had a population of 5,000 people. Remoteness for each locality was then classified based on a score from 0 (high accessibility) to 12 (high remoteness) and grouped into five categories: ’highly accessible’, ’accessible’, ’moderately accessible’, ’remote’ and ’very remote’ services (AIHW, 2004).

In ARIA+ the remoteness index value was based on distance to 5 categories of ’service centre’ that includes centres with populations from 1,000-4,999 to reflect the impact of small centres as localities with populations of greater than 1000 persons were considered to contain at least some basic level of services (for example health, education, or retail) (GISCA, 1999). Those ’service centres’ with larger populations were assumed to contain a greater level of service provision. ASGC remoteness categories were then given to Census Collection Districts (CDs) based on the average ARIA+ score within the CD. This was done at the level of the SLA with ARIA. As CDs are generally smaller than SLAs this provides a greater level of precision of the measure of remoteness. SLAs under ARIA+ are then classified by the proportion of the population living in CDs by the ASGC Remoteness Area classes (AIHW, 2004). The names of the first 3 classes of remoteness -metropolitan, inner regional and outer regional have changed also from those used by ARIA. The names for remote and very remote classes remain the same although they are assigned slightly differently under ARIA+. This has resulted in some change in the proportions of the population classified under the new ARIA+ categories compared to the previous ARIA categories. However, the effect on rural and remote areas appears small. It has also meant that some SLAs have been reclassified. The proportion of the population of each SLA in each ARIA+ category is shown in a separate appendix to this report.

4.3 Country of birth

Where possible, indicators for the five countries that comprise the Former Yugolsavia are presented separately. However, in many datasets these countries are not identified separately, in which case indicators are presented for the Former Yugoslavia. The five countries that comprise the Former Yugolsavia are: Croatia, Slovenia, Bosnia-Herzegovina, the Former Yugoslav Republic of Macedonia, and the Former Yugoslav Republic of Serbia and Montenegro.

4.4 Socioeconomic status

The following relates to methods used in the chapter on socioeconomic status.

4.4.1 Socioeconomic status measures

Socioeconomic (SES) groups used in this chapter were constructed using the index of relative socioeconomic disadvantage (IRSD), which is one of the socioeconomic indices for areas (SEIFA) produced by the Australian Bureau of Statistics (ABS) (ABS, 1998, 2003). Non-overlapping geographical areas covering all of NSW are assigned an IRSD score calculated from ABS census data on various socioeconomic characteristics of the people living in the areas. These characteristics relate to occupation, education, non-English speaking background, indigenous origin, and the economic resources of the household.

The ABS has released IRSD scores after the last four censuses. The methods used for calculation of the IRSD index were similar in 1986, 1991 and 1996, but changed for 2001. The IRSD score is an ordinal measure based on a standard score of 1000 and standard deviation of 100 for Australia, based on the index scores of all collector districts (CDs) in Australia. The areas can be ranked by IRSD score but other arithmetic comparisons using the score are not valid. Only ranks, and not the scores calculated using data from different censuses, can be compared. For instance, the score for NSW was 1006 using 1996 census data, which means that the SES of NSW was slightly better than Australia as a whole. The score for NSW in 1991 was 1002; however that does not mean that NSW in 1996 was better off than NSW in 1991 because the scores were calculated based on a socioeconomically different Australian population. Calculations of the IRSD scores for a local governemnt area involves the weighting of the indexes based on the population for the particular year.

The NSW population was divided into 3 groups for the analyses in this chapter. Statistical local areas were sorted by IRSD score and assigned to quintiles, each containing as close as possible to one-fifth of the total population. The data are presented for the lowest SES population quintile, the highest SES population quintile, and a group comprising the remaining three SES quintiles, which is referred to as the 'rest' or 'balance of NSW population'.

4.4.2 Poisson regression models

Poisson regression models (Armitage et al., 2002) were used to study the effect of time and SES on death rates. For each indicator (except life expectancy and ACS hospitalisations) and each sex, the trend for SES group by time was modelled to obtain fitted values for the relative health gaps and to ascertain the significance of any observed changes in the health gaps over time. Raw ratios were used to assess changes in the relative position of the SES groups for life expectancy or the rate of hospitalisations for ACS conditions due to difficulties in fitting the Poisson model in these cases.

The models included age, SES group, and year and the interaction of year and SES group. The interaction term assessed change in death rates by SES group over time, after adjusting for age differences. The relative rate of change was determined by exponentiating the coefficient for the appropriate SES*time variable in the model. The significance of the change was assessed by testing the difference between the slopes of these trends using the CONTRAST option in the GENMOD procedure in SAS (SAS, 2001).

4.5 Avoidable mortality

The method used to calculate avoidable mortality was based on a revision of the original set of conditions published in 2001 (Tobias and Jackson, 2001). This review, by the Public Health Information Development Unit (PHIDU) in Australia and the Ministry of Health in New Zealand, aims to develop an Australasian standard list of potentially avoidable conditions.

Avoidable deaths are those attributed to conditions that are considered preventable or otherwise avoidable through earlier intervention or action. These were further sub-categorised into 3 levels of intervention. Primary level interventions are those that can prevent the condition developing, such as promotions of lifestyle modification. Secondary level interventions are those that detect or respond to the condition early in its progression, such as cancer screening and chronic disease management. Tertiary level interventions are those that treat the condition and prevent premature death. For each condition, the number of deaths that could have been avoided at each level was calculated by applying weights to the total deaths from the condition. These data were summed to determine the rates of primary, secondary and tertiary avoidable mortality. The weights were based on the work of Tobias and Jackson (2001).

The codes used to define avoidable mortality groups, along with the weights for defining the proportion avoidable by primary, secondary and tertiary interventions can be found in the disease and procedure codes section of this report.

4.6 Avoidable hospitalisations

The method used to calculate avoidable hospitalisations used the concept of ambulatory care-sensitive (ACS) conditions. These are hospitalisations that could have been avoided through the use of preventive healthcare or early disease management given in an ambulatory setting, such as by a general practitioner or community health centre.

The categories used for the ambulatory care-sensitive conditions are based on those used by the Victorian Department of Human Services (DHS, 2001, which have been reviewed by the Public Health Information Development Unit (PHIDU). The DHS list was defined according to ICD-9-CM; appropriate groupings of ICD-10-AM codes were developed for the previous report, and have undergone review by PHIDU.

The information presented in this report differes from information presented in previous reports The health of the people of New South Wales. In 2006 in NSW the coding of diabetes was changed to include diabetes in primary diagnosis only and the coding of cellulitis has been brought in line with the original PHIDU's definition.

The codes used can be found in the disease and procedure codes section of this report.

4.7 Diabetes-related deaths

The term 'diabetes-related death' is used in this report to refer to deaths where either diabetes was recorded and the underlying cause of death, or where diabetes was recorded as an associated cause of death and the underlying cause of death was one of a specific list of commonly recognised diabetes complication. These complications were: myocardial infarction, ischaemic heart disease*, stroke or sequelae of stroke*, heart failure*, sudden death (cardiac arrest), peripheral vascular disease, kidney disease, hyperglycaemia, hypoglycaemia (AIHW, 2005).

The reason for this approach was that, more than other disorders, diabetes often causes death indirectly because it is a strong risk factor for common causes of death such as heart and kidney disease, and stroke. These complications are likely to appear and the underlying cause of death, the basis for official mortality statistics. If only cases where diabetes as the underlying cause were counted, it would lead to considerable underestimates of diabetes' contribution to death in Australia (AIHW, 2005).

The concept of 'diabetes-related deaths' is based on the definition of 'death related to diabetes' used in the United Kingdom Prospective Diabetes Study since 1998. The UKPDS definition has been modified by diabetes specialists on the National Diabetes Data Working Group, associated with AIHW, to include additional conditions (marked above with an asterisk*) (AIHW, 2005).

For the full list of codes of included conditions see the disease and procedure codes section of this report.

4.8 Psychological distress

The K10 (Kessler and Mroczek, 1992, 1994; Kessler et al., unpublished) was included in the 1997, 1998, 2002 and 2005 NSW Population Health Surveys as a relatively short measure of psychological distress that allowed comparison against international survey data and validation against concurrent diagnostic data in the National Survey of Mental Health and Wellbeing (NSMHW) (ABS, 1997; Andrews and Slade, 2001). The K10 (and a briefer K6 measure) were specifically designed for use in the 'core' of the annual US National Health Interview Survey (NHIS, N=50,000 aged 15+) when it was redesigned for use from 1997 onwards. The K6 has also been used in the biennial Canadian National Population Health Survey (panel survey, N>17,000 aged 12+; 1994-95; 1996-97; 1998-99; 2000-01) and has been replaced by the K10 in the Canadian Community Health Survey from September 2000 (Statistics Canada, 2002).

The K10 is currently being used in a series of surveys similar to the Australian NSMHW, in 20 countries, under the auspices of the World Health Organisation (WHO). These surveys have a total sample size of about 200,000. The WHO regions surveyed include North America (Canada and the United States), Latin America (Brazil, Colombia, Mexico and Peru), Europe (Belgium, France, Germany, Italy, The Netherlands, Spain, and The Ukraine), the Middle East (Israel), Africa (South Africa), and Asia (China, India, Indonesia, Japan, and New Zealand) (Kessler et al., 2000).

The K10 measure is a 10-item self-report questionnaire intended to yield a global measure of 'psychological distress' based on questions about the level of restlessness, anxiety, and depressive symptoms in the most recent 4-week period. It is designed to span the range from few or minimal symptoms through to extreme levels of distress, which is an essential feature of an instrument for use in population studies. Thus the K10 contains both low-threshold items, that many people may endorse, through to high-threshold items that very few will endorse. Overall, the item-response scale is designed to yield most precision around the 90th to 99th percentile of the general population.

For each item there is a 5-level response scale based on the amount of time (from none through to all) during a 4-week period when the person experienced the particular problem. In NSW use, there are also 4 follow-up questions, that aim to quantify the level of disability resulting from the feelings of distress; the health service usage resulting from the distress; and the extent to which the distress is believed to be mainly due to physical health problems.

Scoring of the raw questionnaire assigns between 1 to 5 points to each symptom in the direction of increasing problem frequency. Thus, the raw score range is from 10 (all responses to all questions are 'none of the time') through to 50 (all responses to all questions are 'all of the time'). Scoring of the raw questionnaire assigns between 1 to 5 points to each symptom in the direction of increasing problem frequency. Thus, the raw score range is from 10 (all responses to all questions are 'none of the time') through to 50 (all responses to all questions are 'all of the time'). Low scores are indicating low levels of psychological distress and high scores indicating high levels of psychological distress (ABS, 2003).

The creators of the K10 have not yet published details on scoring the scale and there has been no international standard for determining cut off points for low, medium and high levels of psychological distress (ABS, 2003). Various interpretations of scoring were used in the past in Australia and worldwide. Recently, and following the advice of the K10 originators, NSW adopted a four level approach to illustrate prevalence and severity. The four levels are given in Table 2.

Table 2. K10 score and level of psychological distress
K10 score	Level of psychological distress
10-15	Low
16-21	Moderate
22-29	High
30-50	Very high

These cut-off scores were previously used in the 2000 Health and Wellbeing Survey (conducted in Western Australia) and the ABS 2001 National Health Survey Summary of Results Publication (ABS, 2003). The adoption of the above scores in NSW ensures comparability of the NSW results with national and, increasingly, international data.

5. Area Health Service boundaries

In July 2004, the Minister for Health announced a new area health structure. The restructure is largely a merger of the 17 Area Health Services into 8 new Area Health Services. The new administrative structure will be in place by January 1, 2005. A list of the mergers is given in Table 3. Some localities within the Statistical Local Areas of Greater Taree, Greater Lakes, Gloucester and Lithgow moved to the new Area Health Services independently of the main mergers.

In this report, the 2005 Area Health Services boundaries have been used. A list of local government areas by 2005 Health Area is given elsewhere in this appendix.

Table 3. 1996 and 2005 Area Health Services
1996 Area Health Service names	2005 Area Health Service names
Central Sydney South Western Sydney	Sydney South West
South Eastern Sydney Illawarra	South Eastern Sydney & Illawarra
Wentworth Western Sydney	Sydney West
Northern Sydney Central Coast	Northern Sydney & Central Coast
Hunter New England	Hunter & New England
Mid North Coast Northern Rivers	North Coast
Greater Murray Southern	Greater Southern
Far West Macquarie Mid Western	Greater Western
Note: Some localities within the Statistical Local Areas of Greater Taree, Greater Lakes, Gloucester and Lithgow moved to the new Area Health Services independently of the main mergers.

6. Quality assurance process

The preparation of this report involved a great deal of complex data processing and manipulation. The following steps were taken to minimise errors:

· most analyses used a single, shared suite of datasets, contained on the HOIST system. The datasets on HOIST are carefully checked against the original source data to ensure their fidelity. Sources for all data used are described in the footnotes;

· all graphs and tables were produced using SAS programs that can be audited, rather than using interactive data manipulation facilities such as spreadsheets that are much more difficult to check;

· the SAS programs directly created the Web pages for the online version of the report as well as tables and graphs, which were directly imported into Adobe PageMaker for typesetting the printed version. This minimised the possibility of transcription and typographical errors;

· every SAS program used in the production of this report was checked by someone other than the person who originally wrote it. Items such as the correct specification of ICD codes and correct selection of numerator and denominator data were systematically checked as part of this audit process;

· complex parts of the SAS programs were abstracted as a common, shared set of SAS 'macros' (callable subroutines). These macros-which were employed for operations such as imputation, direct standardisation and production of custom graph formats-were subject to rigorous testing before they were used;

· all results were checked against other, comparable sources wherever possible.

Top of page

Table of contents