SEARCH

SEARCH BY CITATION

Keywords:

  • countries;
  • epidemiology;
  • public health;
  • stroke;
  • stroke burden;
  • stroke prevalence

Abstract

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References
  8. Supporting Information

Background

Estimates of strokes in Australia are typically obtained using 1996–1997 age-specific attack rates from the pilot North East Melbourne Stroke Incidence (NEMESIS) Study (eight postcode regions). Declining hospitalizations for stroke indicate the potential to overestimate cases.

Aims

To illustrate how current methods may potentially overestimate the number of strokes in Australia.

Methods

Hospital separations data (primary discharge ICD10 codes I60 to I64) and three stroke projection models were compared. Each model had age- and gender-specific attack rates from the NEMESIS study applied to the 2003 population. One model used the 2003 Burden of Disease approach where the ratio of the 1996–1997 NEMESIS study incidence to hospital separation rate in the same year was adjusted by the 2002/2003 hospital separation rate within the same geographic region using relevant ICD-primary diagnosis codes. Hospital separations data were inflated by 12·1% to account for nonhospitalized stroke, while the Burden of Disease model was inflated by 27·6% to account for recurrent stroke events in that year. The third model used 1997–1999 attack rates from the larger 22-postcode NEMESIS study region.

Results

In 2003, Australian hospitalizations for stroke (I60 to I64) were 33 022, and extrapolation to all stroke (hospitalized and nonhospitalized) was 37 568. Applying NEMESIS study attack rates to the 2003 Australian population, 50 731 strokes were projected. Fewer cases for 2003 were estimated with the Burden of Disease model (28 364) and 22-postcode NEMESIS study rates (41 332).

Conclusions

Estimating the number of strokes in a country can be highly variable depending on the recency of data, the type of data available, and the methods used.


Introduction

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References
  8. Supporting Information

Researchers, clinicians, and policymakers alike require robust estimates of the number of strokes occurring within individual countries. This is to ensure that appropriate policies and decisions are made within an environment of limited resources. For clinicians and healthcare providers, knowing the scale of the stroke burden is also important for addressing workforce capacity and training needs. Unfortunately, as highlighted in this issue of the International Journal of Stroke, many countries do not have high-quality data on stroke, and in countries where the data do exist, it is often quite old and may be inaccurate. Further, because the definitions of stroke events are changing, comparability of stroke attack rates may be unreliable. For example, a tissue-based definition may increase attack rates where cases of ‘silent stroke’ are found. In this report, reference to stroke incidence is based on the classical epidemiological definition of first-ever events using World Health Organization criteria [1].

In Australia, the current estimates for stroke are variable and are reported to range between 35 000 and 60 000 new events per annum depending on whether the estimates are based on data from hospitalizations or community-based epidemiological studies of stroke attack rates undertaken in specific locations and extrapolated to the rest of the country [2-6]. Use of hospitalization separations has the advantage of providing national data, while community-based stroke epidemiological studies include those people with stroke who do not go to hospital (Table 1).

Table 1. Common sources of data used for estimating stroke numbers and their strengths and limitations
Community-based incidence studiesDRG hospital admissions data
  1. DRG, diagnostic related group; ICD, International Classification of Diseases.

Limited generalizability to other regionsNational coverage and so generalizable
Time consuming/costlyEasily obtained
Nonhospitalized events includedExcludes nonhospitalized events
Standard definition of stroke and subtypes

Potential coding inaccuracies of stroke

Stroke subtype coding poor (better if ICD codes available)

Outcome data availableLimited outcome information
Person-level informationEpisode based, not patient based

Although there have been several community-based epidemiological studies of stroke in Australia, the North East Melbourne Stroke Incidence Study (NEMESIS) is the one most often used to extrapolate and estimate the numbers of strokes occurring in Australia [5]. The first wave of NEMESIS (pilot study) was undertaken between 1996 and 1997 using multiple overlapping sources to ascertain all cases of stroke in a population of 133 816 residents in suburbs north and east of Melbourne covering eight postcode regions, and standard definitions for stroke and case fatality were used [5]. In the two subsequent years (1997–1999), ascertainment of stroke cases was obtained for a larger 22-postcode region, but these more recent data have not been used for nationwide projections [7]. For example, the 2003 Burden of Disease (BoD) Study conducted by Begg and colleagues have used NEMESIS incidence rates as part of their stroke models for estimating the BoD in Australia, but with some adjustments for observed temporal declines in stroke hospitalizations since 1997 [6, 8] (Fig. 1).

figure

Figure 1. Projected stroke hospitalizations based on 1996–1997 hospitalizations. Each line represents projections based on a different rate of change in hospitalizations for the financial year 1996–1997, as well as population growth and expected changes in population demographics. The orange line shows observed hospitalizations.

MH-RR, Mantel–Haenszel age-adjusted rate ratio; adapted from Thrift et al. [6].

Download figure to PowerPoint

The aim of this report is to illustrate the potential for wide variability in the estimation of the number of strokes occurring using Australia as a case study. Below we provide four examples of the estimated number of strokes for Australia.

Methods

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References
  8. Supporting Information

We compared four different options for estimating the number of strokes in Australia for a 2003 reference year: (1) hospital separations data (primary diagnosis ICD10 codes I60 to I64) obtained from the Australian Institute of Health and Welfare (AIHW; http://www.aihw.gov.au/data-cube/?id=10737418525) (accessed 16 September 2013) and inflated to include nonhospitalized cases (12·1% from NEMESIS); (2) a stroke projection model from the 2003 BoD study using adjusted pilot NEMESIS incidence rates to account for declines in hospital separations for stroke since 1997 (detailed below) and then adjusted to include recurrent events; (3) a project model based on 1997–1999 attack rates from the larger 22-postcode NEMESIS region (see Supporting Information Table S1); and (4) a model based on projections of stroke numbers using the most commonly used attack rates from NEMESIS (referred to here as the 1996–1997 pilot NEMESIS model). Each projection model was based on age- and gender-specific attack rates from NEMESIS and applied to the 2003 Australian population [9]. Where applicable, it was assumed that the proportion of recurrent events (27·6%) and proportion of nonhospitalized cases (12·1%) were the same between 1997 and 2003.

The BoD model was the one used in the 2003 Australian BoD study [10]. In the BoD model used here, Vos and colleagues multiplied the 2002/2003 age-specific hospital separation rate for primary diagnosis of ICD-10 I60-69, by the ratio of NEMESIS incidence to age- and gender-specific hospital separation rate for primary diagnosis of ICD-9 430-438 in the statistical local areas (SLAs) closely matching the NEMESIS postcodes (SLAs 21891, 21892, 20661, 20662) for 1996/1997. Further, because in NEMESIS there was zero incidence rates for some people aged less than 25 years, this was assumed to be due to small sample sizes. Therefore, it was assumed that the incidence was probably no different between the genders in the lower age groups and that the female incidence was equal to male incidence rates in the 0–14 and 15–24 age groups. The results of the adjusted hospital separations data and the two stroke projection models were compared.

The equations used for each model are:

  • Hospital separations data adjusted to include nonhospitalized stroke: Sum of ICD10 separations for I60 to I64 × 12·1% for nonhospitalized cases
  • Adjusted BoD model = 1996–1997 NEMESIS incidence rates (age/gender) × 2003 hospitalization for ICD10/1997 hospitalization for ICD9 within the NEMESIS geographic areas. Then this result was adjusted to include recurrent events (with 27·6% more cases)
  • 22-postcode NEMESIS model = 1997–1999 attack rates (age/gender) × Australian population (2003)
  • Pilot region NEMESIS model = 1996–1997 NEMESIS attack rates (age/gender) × Australian population (2003)

Results

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References
  8. Supporting Information

In 2003, the national population was estimated to be 19 872 646, of which 49·7% were male and 13% were aged 65 + years. Australian data from the AIHW on hospitalizations for stroke were 33 022, and when adjusted to include nonhospitalized cases was estimated to be 37 103 strokes (Fig. 2). The estimated numbers of stroke in 2003 unadjusted for the declining rate of stroke using 1996–1997 pilot region NEMESIS attack rates were 50 731. When adjusted for the declining hospitalization rate using the BoD model, the estimated number of strokes was 28 364, a 44% reduction. The model using the 22-postcode NEMESIS region attack rates provided an estimate for 2003 of 41 332 cases. The BoD model had about 24% fewer strokes than the AIHW figures, while the 1996–1997 pilot NEMESIS model had 26% more strokes than the figures estimated using AIHW data. The 22-postcode NEMESIS region model had 9% more strokes than the hospitalizations data from the AIHW, while it had 19% fewer stroke than the1996–1997 pilot model.

figure

Figure 2. Summary estimations of stroke numbers for Australia in 2003 using different methodological approaches and data sources.

AIHW, Australian Institute of Health and Welfare; NEMESIS, North East Melbourne Stroke Incidence study.

Download figure to PowerPoint

Discussion

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References
  8. Supporting Information

We found wide variability in the estimates of strokes occurring in Australia, ranging from 28 364 to 50 731. The population in Australia is rapidly changing, with an increasing proportion of people aged 65 years or more. This means that the number of strokes occurring would be expected to increase even if stroke attack rates remain the same over time. There is mixed evidence of declines in the number of estimated first-ever cases of stroke in Australia and New Zealand, indicating that this may range from 2% to 5% since the 1990s [11-13]. Recent data on stroke incidence in South Australia provide evidence that stroke incidence has declined when contrasted to the earlier Australian studies (adjusted to world population South Australia 76 per 100 000 [95% CI 59 to 94] compared with NEMESIS pilot region 100 per 100 000 [95% CI 80, 119]) [5, 14]. However, it appears that the mix of stroke subtypes is changing as cardioembolic strokes were a larger proportion of all ischemic stroke than in the other studies [14-16].

The estimates for stroke events in Australia are variable and are the result of imperfect data. We have illustrated that methods of adjusting estimates to account for declines in hospital separations (as per the BoD model) or the reliance solely on stroke hospital separations data based on primary diagnosis codes (which may, for example, exclude strokes that occur in hospital) may underestimate stroke events. In contrast, the use of outdated attack rates from the 1996–1997 pilot NEMESIS region, which were obtained from a geographical region that has a more socioeconomically disadvantaged population compared with other regions of Australia, may overestimate events. The estimates from the 22-postcode NEMESIS region reflected a more consistent picture with the 2003 hospitalizations data, and this may be partly explained because these data were obtained from a population where the socioeconomic profile was better matched to the rest of Australia [7]. The reason these attack rates have not been used as often for projecting national stroke estimates is because full information on the attack rates have not been published (see Supporting Information Table S1). In our opinion, the true number of strokes lies somewhere in between the model using the estimates from the 22-postcode NEMESIS region and the model using the hospital separation data. Ideally, this range should be quoted and adopted in future work.

Data used to estimate the numbers of stroke in a country can result in highly variable numbers which may adversely impact policy or health system planning for stroke. Community-based stroke epidemiological studies provide an ‘ideal’ approach to estimating stroke numbers because they include hospitalized and nonhospitalized cases, but the region used should reflect the socioeconomic profile of the relevant country. More efficient methods of obtaining community-based epidemiological data are needed as these projects are expensive and labor intensive. This study highlights the need to agree on standard methods for adjusting estimates where outdated data may be the ‘best available’ data and the importance of verifying figures against other sources of data. Lastly, where possible community-based stroke epidemiological studies should be repeated so that better data are available for one of the largest causes of disease burden and this also permits validation of case ascertainment methods.

References

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References
  8. Supporting Information
  • 1
    Hatano S. Experience from a multicentre stroke register: a preliminary report. Bull World Health Organ 1976; 54:541553.
  • 2
    Cadilhac DA, Carter RC, Thrift AG, Dewey HM. Why invest in a national public health program for stroke? An example using Australian data to estimate the potential benefits and cost implications. Health Policy. 2007; 83:287294.
  • 3
    Deloitte Access Economics. The Economic Impact of Stroke in Australia. Sydney, Deloitte Access Economics, 2013:13 March. Report No.
  • 4
    Senes S. How We Manage Stroke in Australia. AIHW cat no CVD 31. AIHW, editor. Canberra, Australian Institute of Health and Welfare, 2006.
  • 5
    Thrift AG, Dewey HM, Macdonell RA, McNeil JJ, Donnan GA. Stroke incidence on the east coast of Australia: the North East Melbourne Stroke Incidence Study (NEMESIS). Stroke 2000; 31:20872092.
  • 6
    Thrift AG, Tong B, Senes S, Waters AM, Lalor E. No evidence for an epidemic of stroke with the ageing of the population. Neuroepidemiology 2012; 38:268273.
  • 7
    Thrift AG, Dewey HM, Sturm JW et al. Greater incidence of both fatal and nonfatal strokes in disadvantaged areas: the Northeast Melbourne Stroke Incidence Study. Stroke 2006; 37:877882.
  • 8
    Begg SJ, Vos ET, Barker B, Stevenson CE, Stanley L, Lopez AD. The Burden of Disease and Injury in Australia 2003. PHE 82. Canberra, Australian Institute of Health and Welfare, 2007.
  • 9
    Australian Bureau of Statistics. Australian Population Projections Series B 2002–2101. Canberra, Australian Bureau of Statistics, 2005.
  • 10
    Begg SJ, Vos T, Barker B, Stanley L, Lopez AD. Burden of disease and injury in Australia in the new millennium: measuring health loss from diseases, injuries and risk factors. Med J Aust 2008; 188:3640.
  • 11
    Anderson CS, Carter KN, Hackett ML et al. Trends in stroke incidence in Auckland, New Zealand, during 1981 to 2003. Stroke 2005; 36:20872093.
  • 12
    Islam MS, Anderson CS, Hankey GJ et al. Trends in incidence and outcome of stroke in Perth, Western Australia during 1989 to 2001: the Perth Community Stroke Study. Stroke 2008; 39:776782.
  • 13
    Marsden DL, Spratt NJ, Walker R et al. Trends in stroke attack rates and case fatality in the Hunter region, Australia 1996–2008. Cerebrovasc Dis 2010; 30:500507.
  • 14
    Leyden JM, Kleinig TJ, Newbury J et al. Adelaide stroke incidence study: declining stroke rates but many preventable cardioembolic strokes. Stroke 2013; 44:12261231.
  • 15
    Anderson CS, Jamrozik KD, Burvill PW, Chakera TM, Johnson GA, Stewart-Wynne EG. Determining the incidence of different subtypes of stroke: results from the Perth Community Stroke Study, 1989–1990. Med J Aust 1993; 158:8589.
  • 16
    Thrift AG, Dewey HM, Macdonell RA, McNeil JJ, Donnan GA. Incidence of the major stroke subtypes: initial findings from the North East Melbourne stroke incidence study (NEMESIS). Stroke 2001; 32:17321738.

Supporting Information

  1. Top of page
  2. Abstract
  3. Introduction
  4. Methods
  5. Results
  6. Discussion
  7. References
  8. Supporting Information
FilenameFormatSizeDescription
ijs12246-sup-0001-si.doc32K

Table S1. Standardised attack rates from the North East Melbourne Stroke Incidence Study 22 post-code region using data obtained from 1997 to 1999.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.