Increased chronic disease prevalence among the younger generation: Findings from a population-based data linkage study to inform chronic disease ascertainment among reproductive-aged Australian women.

Chronic disease represents an ongoing public health challenge. It contributes substantially to global healthcare expenditure and disease burden in terms of health service use, hospitalisations as well as costs associated with treatment regimens [ 1 , 2 ]. Australia, in particular, is facing a chronic disease crisis. In 2011, chronic disease was reported as an underlying cause in 90% of deaths [ 3 ]. Despite the increased focus on preventative health strategies, chronic disease rates continue to rise with estimates from the 2017–18 National Health Survey (which provides a snapshot of Australia’s health) suggesting that half of adults have at least one chronic health condition (up from one-third in 2007–08) [ 1 , 4 ]. Women in particular appear to be disproportionately affected, and at younger ages compared to men [ 5 ]. Of the common conditions listed in the recent National Health Survey, 26% of women reported one chronic disease and 23% reported two or more conditions (multimorbidity). Importantly, while 83% of Australian women reported chronic disease when aged over 65 years, 43% reported chronic disease during their reproductive years (i.e., <45 years). Findings from population-based cohort studies support the increasing prevalence of chronic diseases such as diabetes across successive generations [ 6 ]. Accurate prevalence and ascertainment of chronic disease among women of reproductive age at the population level is essential for not only public health surveillance, policy and healthcare planning but also meeting the family planning and reproductive health challenges that chronic diseases pose [ 7 ]. While a number of data sources have been used for monitoring chronic disease trends such as population-based surveys, medical chart review, disease registries and administrative healthcare databases (e.g., primary health care, hospitalisation and pharmaceutical), there is no clear reference standard for the ascertainment of chronic disease. Epidemiological research focused on ascertainment techniques among older populations have demonstrated that a single data source is insufficient to accurately capture chronic diseases. All data sources have limitations (with variable agreement between sources depending on the disease of interest), which may contribute to unreliable case ascertainment and biased prevalence and incidence estimates [ 8 – 14 ]. Even in diseases such as diabetes, which can be clearly identified in administrative data, disease prevalence and incidence may still be under-reported. For example, in the case of type 2 diabetes, one-third of cases are controlled by diet and lifestyle alone leading to fewer diabetes specific records in administrative data [ 15 ]. Therefore, in order to increase the validity of chronic disease ascertainment, use of multiple data points through linkage of different data sources (including the development of tailored algorithms on a disease specific basis) is increasing [ 11 , 16 – 22 ]. In Australia, chronic disease is managed through a tax-funded universal healthcare system, which provides researchers access to rich and long-term health service and pharmaceutical claims data. Additionally, the Australian Government has funded the Australian Longitudinal Study on Women’s Health (ALSWH) since 1996, to follow a cohort of women born 1973–78 through their reproductive years, over two decades, collecting information on their health outcomes and use of health services, as well as social, behavioural and economic factors which influence health. This national longitudinal study also recruited a new cohort in 2012, aiming to collect similar information on the next generation of younger Australian women [ 23 , 24 ]. As these data are routinely linked to administrative health records, there is an ideal opportunity to estimate the prevalence of chronic diseases among women of reproductive age using multiple data sources, and to assess the concordance between available data sources. This approach has not previously been undertaken specifically for women of reproductive age. With chronic disease an ongoing public health issue and with chronic disease development occurring earlier in life, particularly for women, it has implications for the delivery of family planning support strategies and preconception contraceptive counselling as well as policy implications for advancing reproductive health. The aim of this paper is to estimate the prevalence of chronic disease among younger Australian women of reproductive age, in order to identify conditions that would benefit from targeted family planning support strategies and preconception contraceptive counselling interventions.

The 1973–78 cohort who completed the baseline survey in 1996 contained 14,247 women, of whom 13,501 (94.8%) did not opt out of data linkage and were eligible for inclusion in this analysis (see Fig 2 ). Similarly, only a small percentage (0.3%) of the 17,010 women from the 1989–95 cohort did not provide consent to data linkage, leaving 16,964 women eligible for analysis. Baseline characteristics for women included in the analysis are shown in S4 Table and were obtained from the first survey completed, either in 1996 (1973–78 cohort) or 2012/2013 (1989–95 cohort). Most women in both cohorts were Australian born, unpartnered and lived in a major city at their baseline survey. A higher proportion of women from the 1989–95 cohort were overweight or obese (32.5% vs 21.9%), had achieved tertiary qualifications (21.9% vs 11.1%), were non-smokers (81.2% vs 52.1%) and rated their health as fair or poor (16.9% vs 12.2%) compared to the 1973–78 cohort. The number and types of data sources used to ascertain cases varied across chronic diseases ( Table 2 ). Between 28% and 70% of chronic diseases were ascertained through one data source only, depending on the condition. Diabetes had the highest prevalence of ascertainment through a single source (70%) while mental health conditions had the lowest prevalence (28%). Diabetes, autoinflammatory arthropathies and connective tissue disease, inflammatory bowel disease, and chronic kidney disease were the most likely to be ascertained through administrative data only, with 68% to 79% of cases identified in this way. However, the prevalence of each condition within each administrative data source differed depending on the condition. Diabetes was most frequently ascertained through the MBS, autoinflammatory arthropathies and connective tissue disease through the PBS and inflammatory bowel disease through the APDC. Multiple sclerosis had the highest proportion of cases identified through survey only, while mental health conditions, inflammatory bowel disease, and cancer had the highest proportion of cases identified through three or more data sources. a Eligible women from the Australian Longitudinal Study on Women’s Health (born 1973–78 or 1989–95). CKD = Chronic Kidney Disease; Arthrop = Autoinflammatory arthropathies and connective tissue disease; IBD = Inflammatory bowel disease; MS = Multiple Sclerosis. MBS = Medicare Benefits Schedule; PBS = Pharmaceuticals Benefits Scheme; APDC = Admitted Patient Data Collection; NDI = National Death Index; ACD = Australian Cancer Database. * Across all conditions, only 68 indications were obtained from NDI, as there were only 128 deaths across the two cohorts during the observation period. Low concordance between data sources was reported for all examined conditions (see Table 3 ), regardless of the sources examined. Survey and PBS had the highest concordance for several conditions, although the degree of concordance varied widely (from 5% for chronic kidney disease up to 46% for asthma). Overall, the highest concordance was found between the APDC and PBS for inflammatory bowel disease (49%). a Eligible women from the Australian Longitudinal Study on Women’s Health (born 1973–78 or 1989–95). CKD = Chronic Kidney Disease; Arthrop = Autoinflammatory arthropathies and connective tissue disease; IBD = Inflammatory bowel disease; MS = Multiple Sclerosis. APDC = Admitted Patient Data Collection; PBS = Pharmaceuticals Benefits Scheme; MBS = Medicare Benefits Schedule; ACD = Australian Cancer Database. Figs 3 and 4 show the yield gain in chronic disease prevalence for each of the cohorts with the use of additional data sources (based on the typical clinical disease course) over the observation period. The highest yield gain for asthma was the addition of survey self-report to the MBS for asthma (increasing from 2.4% to 11.6% for the 1989–95 cohort and 1.8% to 16.7% for the 1973–78 cohort). It also provided the highest yield gain for cardiac disease (increasing from 0.4% to 1.4% for the 1989–95 cohort and 0.6% to 6.1% for the 1973–78 cohort). Gain for asthma and cardiac disease was modest with only a small yield gained by the addition of the PBS and APDC. The highest yield gain for autoinflammatory arthropathies and connective tissue disease included the addition of PBS to survey data with the prevalence increasing from 0.3% to 1.5% for the 1989–95 cohort and 1.6% to 4.3% for the 1973–78 cohort. Only a 0.1% increase in prevalence was demonstrated across both cohorts with the addition of the APDC/NDI datasets. A similar pattern was identified for thyroid disease. Further, the addition of any dataset to the MBS only resulted in small gains in the lifetime prevalence of mental health conditions. Rare conditions such as inflammatory bowel disease, cancer and chronic kidney disease benefitted from the addition of all available data sources with the prevalence incrementally increasing with the addition of each data source. Table 4 shows the cumulative prevalence for the 1989–95 and 1973–78 cohorts by June 2020 when the women were aged 25–31 years and 42–47 years respectively using the developed algorithms. All conditions had a higher lifetime prevalence among the 1973–78 cohort compared to the 1989–95 cohort except for mental health conditions which was estimated at 70.4% for the 1989–95 cohort and 57.6% for the 1973–78 cohort. Chronic kidney disease was rare across both cohorts with less than 0.5% of women reporting the condition. The cumulative prevalence of inflammatory bowel disease however was similar in both cohorts with 0.9% of women from the 1973–78 cohort and 0.7% of women from the 1989–95 cohort reporting these conditions. Overall, a slightly higher proportion of women from the 1989–95 cohort reported having at least one chronic health condition (75.2% vs 71.3%), although a higher proportion of women from the 1973–78 reported multimorbidity (2 or more conditions; 27.4% vs 17.9%). a The standard case ascertainment algorithm requires two of more self-reports through different surveys; the modified algorithm allows for women to self-report once only if it was first reported at their last returned survey. b Cardiometabolic disease is the combined prevalence of diabetes, cardiac disease and chronic kidney disease. c Autoinflammatory disease is the combined prevalence of autoinflammatory arthropathies and connective tissue disease, inflammatory bowel disease, multiple sclerosis and thyroid disease. CKD = Chronic Kidney Disease; Arthrop = Autoinflammatory arthropathies and connective tissue disease; IBD = Inflammatory bowel disease; MS = Multiple Sclerosis. Examining the prevalence of each of the conditions over the observation period, there was a consistent upward trend for cardiac disease, diabetes, and autoinflammatory arthropathies and connective tissue disease across both cohorts (see Figs 5 and 6 ). There were sharp increases in the prevalence of thyroid disease around 2012/13 in both cohorts, although this steep increase persisted for longer among the 1989–95 cohort. The trend in the prevalence of asthma was stable over time, with a slight increase beginning in 2002 for the 1989–95 cohort when they were aged 7–13 years. An increase in the identification of mental health conditions were noted in 1999/2000 for the 1973–78 cohort when items regarding diagnosed depression and anxiety were introduced in the ALSWH Survey. Mental health conditions were identified much earlier among the 1989–95 cohort, with the prevalence beginning to increase rapidly in approximately 2008 (when aged 13–19 years). Comparing chronic disease prevalence between the cohorts when women were approximately the same age (women in the 1989–95 cohort were aged 25–31 in 2020, while women in the 1973–78 cohort were aged 25–30 in 2003), women from the 1989–95 cohort had more than double the prevalence of diabetes to women from the 1973–78 cohort (4.5% vs. 1.3%), with the difference between cohorts in terms of lifetime mental health conditions substantially higher (70.4% vs. 23.6%). Women from the 1989–95 cohort had lower prevalence of multiple sclerosis and cancer compared to women from the 1973–78 cohort at the same age. Prevalence rates were similar between the cohorts for asthma, autoinflammatory arthropathies and connective tissue disease, and chronic kidney disease. However, women from the 1989–95 cohort had higher prevalence of inflammatory bowel disease. In terms of multimorbidity, a consistent upward trend was identified for both cohorts, although this trend appears to be more pronounced in the 1989–95 cohort. The prevalence of multimorbidity by age 25–31 years for the 1989–95 cohort was 17.9%, while multimorbidity for the 1973–78 cohort was only 9.1% at the same age. Table 4 also compares the prevalence rates obtained using the standard algorithms and the modified algorithms for each condition. The standard case ascertainment algorithm requires two or more self-reports through different surveys, whereas the modified algorithm also allows for women to self-report once only if it was first reported at their last returned survey. We observed an increase in the prevalence in all conditions in both cohorts, but this increase ranged from an additional two cases for chronic kidney disease in the 1989–95 cohort, to an increase of 518 asthma cases in the 1989–95 cohort, which increased the prevalence from 14.9% to 18.0%. A similar increase in the prevalence of asthma was observed in the 1973–78 cohort. Overall, the modified algorithm resulted in a negligible increase in the prevalence rates for inflammatory bowel disease and chronic kidney disease and prevalence increases of less than 4% for the other conditions.

Population-level survey data linked to administrative data provides a powerful basis for chronic disease surveillance and allows for the examination of issues pertinent to women with chronic disease who are of reproductive age (e.g., in relation to family planning and pregnancy). Concordance of disease prevalence between multiple data sources in our study was observed to be low, signalling that the use of a single data source alone would be insufficient to capture the true prevalence of the disease. When unable to use multiple data sources for research involving women of reproductive age, we recommend that researchers be explicit about the limitations of the data sources used. Overall, using a multiple method approach we found that conditions such as diabetes, cardiac disease and inflammatory conditions are increasing across both cohorts but at higher rates among younger women. Lifetime mental health conditions were also found to be at alarmingly high levels for women of both cohorts. In order to reduce the increasing chronic disease and multimorbidity trend among women, strategic chronic disease interventions are required to be implemented in childhood and adolescence to ensure the long-term health of not only current but also future generations.

Chronic disease cases among reproductive-aged women were ascertained using data collected from women in the 1973–78 and 1989–95 cohorts of the ALSWH, a national population-based study of health and wellbeing among Australian women. Women from the 1973–78 cohort were randomly sampled through the Medicare Australia database in 1996 when aged 18–23 years (N = 14,247), with women from rural and remote areas sampled at twice the rate as those from urban areas [ 23 , 25 ]. These women have been surveyed on a roughly three-year rolling schedule with follow-up surveys administered in 2000 (Survey 2), 2003 (Survey 3), 2006 (Survey 4), 2009 (Survey 5), 2012 (Survey 6), 2015 (Survey 7) and 2018 (Survey 8). ALSWH recruited a new cohort of eligible women aged 18–23 years in 2012–2013 (N = 17,010) who completed annual online health surveys until 2017, and a sixth survey in 2019. To meet the needs of this generation, the newest ALSWH cohort was recruited through an open recruitment strategy involving a mix of online and offline methods [ 24 ]. Both cohorts of women have been found to be largely representative of women in their age group, with a slight over-representation of Australian-born and tertiary educated women [ 25 , 26 ]. For the majority of women in these cohorts, survey data are available for linkage to a number of health administrative datasets. For the purposes of this study, ALSWH survey data were linked to the following health administrative databases: state and territory-based Admitted Patient Data Collections, National Death Index, Pharmaceutical Benefits Scheme, Medicare Benefits Schedule, and the Australian Cancer Database. The relationships between these data sources are presented in Fig 1 . Data on hospitalisations were obtained from individual state and territory-based datasets which provide information on all public and private admitted patient services provided for each resident, with the exception of the Australian Capital Territory [ACT], Northern Territory [NT] South Australia [SA] and Tasmanian [TAS] collections which only cover nominated public hospitals. These databases contain information on, but are not limited to, admission and separation dates, procedures undertaken as well as diagnoses related to the admission. Diagnoses are coded according to the Australian Modification of the International Statistical Classification of Diseases and Related Problems (ICD-AM) for a primary diagnosis and up to an additional 54 secondary diagnoses (depending on the state and coding version). APDC records were linked to ALSWH participants via a master linkage key using probabilistic linkage methods for the following states/territories and time periods: a) June 2001 to March 2019 for New South Wales (NSW); b) July 2004 to June 2018 for the ACT; c) May 2007 to December 2017 for Queensland (QLD); d) May 1993 to June 2018 for Victoria (VIC); e) July 2001 to March 2018 in South Australia (SA); f) January 1970 to December 2017 in Western Australia (WA); g) January 2007 to December 2017 in TAS; and h) December 2000 to July 2018 in the NT. Causes of death and date of death were determined from the National Death and Morbidity Indexes based on probabilistic matching of person level information such as name, address, gender, state, and date of birth for the period January 1996 to November 2017 [ 27 ]. The PBS database was used to identify ongoing prescription medications for chronic disease management related to each of the conditions of interest. The PBS provides Commonwealth subsidised prescriptions medications for eligible Australian citizens and residents as well as foreign visitors covered by a Reciprocal Health Care agreement. Prior to April 2012 only prescriptions which carried a government subsidy and were above co-payment were recorded. From April 2012, all above and below co-payment government subsidised medications were captured through the PBS database. PBS data for ALSWH participants were deterministically linked using the individuals’ unique Medicare number [ 28 ]. PBS data were included for the period May 2002 to June 2020. The MBS database provides information regarding claims for medical services by Australian citizens, permanent residents and foreign visitors covered by a Reciprocal Health Care agreement as part of Australia’s tax-funded universal health care system. The MBS database provides comprehensive information regarding a range of services, diagnostic procedures and tests provided by general practitioners, psychologists, allied health or medical specialists. MBS data were linked in a similar manner to PBS data for the period January 1996 to June 2020. The ACD is a national compilation of data from all state and territory-based cancer registries which are routinely collected under legal mandate. As a result, this database provides near complete national information on malignant neoplasms (i.e., excluding non-melanotic skin cancer) recorded from public and private hospitals (including inpatient admissions and day stay procedures), radiation oncology units, residential aged care, pathology laboratories, and outpatient units. The ACD contains pertinent information regarding cancer type, date of diagnosis, age at diagnosis, topography and morphology, tumour size as well as information surrounding survival [ 29 ]. Cancer information for ALSWH participants were probabilistically linked for the period January 1982 to December 2016 for all states and territories (except for NT, which is only linked to 2015). Chronic diseases were identified following consultation with the literature on adverse maternal and foetal outcomes, Australia’s national health priority areas, the UK medical eligibility for contraception and the ability to ascertain the cases in the available datasets [ 30 – 32 ]. As the premise of the study was to ascertain key chronic health conditions that would benefit from preconception contraceptive counselling interventions, chronic diseases such as polycystic ovary syndrome and endometriosis were excluded from the chronic disease definition due to their impact on fertility [ 33 , 34 ]. As such, the final set of conditions to be ascertained included: i) diabetes; ii) cardiac disease; iii) chronic kidney disease; iv) asthma; v) autoinflammatory arthropathies and connective tissue disease; vi) inflammatory bowel disease; vii) multiple sclerosis; viii) thyroid disease; ix) mental health conditions; and x) cancer. Where possible, both congenital and acquired forms of the disease were included. The definition for cardiac disease was based on the Faculty of Sexual and Reproductive Healthcare Clinical Guidance Report and included conditions such as conduction disorders, dysrhythmias, congestive heart failure, cardiomyopathy, ischaemic heart disease, rheumatic heart disease, heart defects and pulmonary circulation disorders. Vascular diseases such as cerebrovascular disease, peripheral vascular disease and venous thromboembolism were considered in the context of cardiac complications only and as such were excluded [ 35 ]. Hypertension, a pre-cursor to acquired heart disease, is increasing among people under the age of 50 years largely due to increasing rates of obesity. Hypertension (uncomplicated and complicated forms) was included with cardiac disease due to its impact on maternal and foetal outcomes [ 36 ]. Mental health conditions were focused on the following disorders: mood (e.g., major depressive disorder, bipolar disorder), anxiety and stress (e.g., panic disorder, obsessive compulsive disorder), psychotic (e.g., schizophrenia), personality (e.g., borderline personality disorder) and eating (e.g., anorexia nervosa). All malignant forms of cancer were included (i.e., excluding in situ cancers of the breast and skin). Individual algorithms were developed for each chronic disease of interest, using the available linked datasets. ICD-AM codes supplied for primary and secondary diagnoses in the inpatient hospital datasets were used to identify chronic conditions of interest. Due to the time frame of the data coverage from the linked data sources, both ICD-10AM and ICD-9-AM codes were used for case ascertainment, as appropriate for each data source. For each chronic condition, ICD codes from the Charlson and Elixhauser indices [ 37 – 39 ] were used for case ascertainment with further refinement after consultation with clinical experts for each condition (see S1 Table ). Deaths related to the conditions of interest were identified in a similar fashion to those used for inpatient hospital records (APDC). For PBS data, case identification was performed using the World Health Organization’s Anatomical and Therapeutic Chemical Classification (ATC) codes [ 40 ]. Medications which are typically indicated in the treatment of the selected conditions were identified using the Australian Medicines Handbook, with confirmation from clinical disease experts [ 41 ]. Using the ALSWH survey data, case ascertainment of chronic diseases was based on participants’ self-report of diagnosis or treatment of the conditions of interest. The time frame in which the women were asked to report this information varied by survey and cohort, with conditions containing sub-types (e.g., diabetes and mental health conditions) being asked with increasing precision over time (see S2 Table ). While direct questions about more prevalent chronic conditions were included in questionnaires over time (e.g., diabetes, heart attack), information concerning other conditions of interest may have been provided by the participant themselves in the free-text comments area when asked to specify any other major illness not listed. As a result, the free-text comments were also used for case ascertainment, using a search on keywords (including colloquial and misspelled variants) based on the ICD codes for each of the disease groups (e.g. diab*, daib*, insulin, kidney, crohns, etc). In general, for inclusion as a chronic disease ‘case’, all conditions with the exception of cancer required either: (a) one indication in either the APDC, MBS or NDI; or (b) self-report in two ALSWH surveys; or (c) two or more medications prescribed within a 12-month period, reported in two separate calendar years (as it was assumed that a chronic condition would require regular ongoing medications). Where a medication had multiple indications for treatment (e.g., metformin [A10BA02] is primarily prescribed in the treatment of diabetes but can also be prescribed in the treatment of PCOS), the medication was included if there was an indication for that condition from another data source (e.g., was also captured through MBS data or via survey self-report). For conditions which include transient gestational/antenatal forms, pregnancy-specific ICD 10 codes and medications prescribed during times of pregnancy (based on each child’s date of birth) were excluded from the chronic disease definition. For medication use, this included a 40-week exclusion period for diabetes and depression/anxiety and 20-weeks for gestational hypertension [ 42 ]. Where MBS data were available, one item related to the management of the conditions (and not purely diagnostic) was required for inclusion (with the exception of diabetes). An example algorithm for diabetes is shown in Table 1 , with case ascertainment algorithms for the remaining conditions using multiple data sources provided in S3 Table . Notes: ALSWH = Australian Longitudinal Study on Women’s Health; MBS = Medicare Benefits Schedule; PBS = Pharmaceuticals Benefits Scheme; COD = Cause of Death (from the National Death Index); APDC = Admitted Patient Data Collection. *The MBS Service Incentive Program for diabetes care requires at least one HbA1c measurement per year. It is recommended that HbA1c testing is done every 6 months if meeting targets, or every 3 months if targets are not being met or if therapy has changed. Therefore, at least 2 HbA1c tests within a 12-month period was required for inclusion. ç PBS: Excluding times of pregnancy (defined as 40 weeks), based on child’s dates of birth. This was to exclude women who may have been diagnosed & treated for gestational diabetes. All data for this project were obtained from the ALSWH (see www.alswh.org.au/ for further details), approved under their Expression of Interest process (EOI A696) and provided in de-identified form. This project has ongoing ethical clearance from the University of Newcastle and University of Queensland’s Human Research Ethics Committees. Ethical approval for linkage of ALSWH survey data to the Admitted Patients Data Collections (APDC) was received from the NSW Population and Health Services Research Ethics Committee and other equivalent state and territory-based committees. Linkages to the National Death Index (NDI) and Australian Cancer Database (ACD) were approved by the Australian Institute of Health and Welfare Ethics Committee. Women provided explicit written consent to participate in the ALSWH as well as for linkage to de-identified administrative health records prior to 2005. From 2005, an ‘opt-out’ consent process was approved by the data custodians and relevant ethics committees for data linkage, with participants regularly reminded of this process. ALSWH participants who decline health record linkage are excluded from data linkage requests. The number of women with each condition of interest was ascertained using multiple data sources and the applicable algorithm. Concordance between data sources was calculated as the percentage agreement between each pair of sources among women with the specified condition. For example, if 100 women overall were determined to have diabetes, and 75 of those women had indications from both MBS and PBS, then the percentage agreement between MBS and PBS for diabetes was 75%. Given that the case ascertainment algorithms provided a conservative measure of disease ascertainment using surveys (i.e., a minimum of two separate self-reports were required across surveys), a sensitivity analysis was conducted for each condition to examine the potential impact on disease prevalence from survey loss-to-follow-up. In modified algorithms for the sensitivity analyses, women were included as having a self-reported indication for the condition if the condition was reported at a single survey only which was also the woman’s last completed survey (i.e., they did not have a later opportunity to self-report via survey). Differences between the standard algorithm and the modified algorithm for each condition were compared. All data preparation and analyses were conducted in Stata 15.1 (College Station, TX: StataCorp LLC).

(DOCX) Click here for additional data file. (DOCX) Click here for additional data file. (DOCX) Click here for additional data file. (DOCX) Click here for additional data file.