Introduction

Age is commonly associated with decline in physical functioning, with many older people experiencing decreased ability to maintain independence in their activities of daily living and instrumental activities. Comprehensive health assessment with appropriate follow-up of identified needs is one option for prevention of functional decline and other adverse events among older people, and improving their quality of life [1–3]. The basis for these assessments is that older people experience many preventable problems that would otherwise go undetected, and that small improvements in lifestyle, support and clinical care may result in major functional or quality of life gains across the course of later life [2,3]. The evidence-base for the effectiveness of such health assessments is largely from studies in Europe and the United States of America, with mixed evidence of effectiveness. However, the majority of trials demonstrate improvement in health status for those who have assessments [1,3–5].

In their systematic review of 18 trials, Stuck et al. [3] found that multidimensional assessment with follow-up resulted in less functional decline. There was also a significant protective effect on mortality for people aged in their 70s but a non-significant effect for those in their 80s. Likewise, assessments were more effective among the subset of the trial populations who had lower mortality rates than among those with higher mortality risk. Similar findings were identified by Beswick et al. [1] who reviewed 28 trials of geriatric assessment for people in the community, and 24 trials where assessments were targeted to frail older people. The pooled effect of these trials showed better physical function among people who had assessments, particularly when assessments were generally targeted to people in the community. However, in this review, there was no clear evidence that assessments reduced mortality rates [1].

Another systematic review [6], examined comprehensive geriatric assessments as one component of non-pharmacological interventions for community dwelling older people. The authors identified 15 studies which included assessment, across a pooled sample of 4603 people.

The assessments evaluated in these studies generally involved medical, psychosocial or functional assessments to inform care plans for primary care and rehabilitation. Only three studies demonstrated statistically significant effects of the health assessments [7–9] and overall, the effects were small or negligible. Most of these studies were aimed at apparently well older people in the community with low rates of disability, and the authors consider the possibility that there was little opportunity to detect differences in disability rates across the study periods [6].

The overall consensus from these studies is that health assessments should have potential to improve outcomes for older people, however there are very mixed results across multiple studies and reviews involving different populations and approaches to the assessments and their follow-up. In this study, we used observational data to examine the effects of assessments across a population of older adults, within the context of the Australian health care system. The Australian government has been subsidising annual assessments for people aged 75 years or over since 1999. These assessments are designed to enhance the capacity of primary health care to improve the health and quality of life of older people with complex care needs. They provide an opportunity for in-depth assessment of the person’s health and allow preventive health care and education. Specific aspects of the assessment include patient’s blood pressure; pulse rate and rhythm; medications; continence; immunisation status for influenza, tetanus and pneumococcus; physical function, including the patient’s activities of daily living, and whether or not the patient has had a fall in the last three months; psychological function, including cognition and mood; oral health; nutrition; and social function, including the availability and adequacy of paid and unpaid help, and whether the person is caring for another person.

The assessments could potentially also increase case finding and older person’s access to health services, enabling greater collaboration between general practitioners (GPs), nurses and other health professionals [10,11].

A randomised controlled trial showed that health assessments for older people in Australia were associated with small improvements in quality of life outcomes, but there was no significant difference in the probability of hospital admission or death [12].

Since their introduction in 1999, the health assessments have been widely used by older people across Australia [13–15], with evidence of their impact on the identification and management of health problems [16–18]. However, our previous research into the uptake of health assessments by older Australian women shows that those who have health assessments are already having more GP visits, and also that few women have repeat assessments as intended by the program [14].

The objective of the current study was to undertake a long-term evaluation of the effect of the health assessments on survival and physical functioning, by comparing outcomes for older women who have had at least one health assessment with those who have never had an assessment, using propensity score matching. We hypothesised that having at least one health assessment after the age of 75 leads to better survival and improved physical functioning. We also expected that health assessments would be associated with higher levels of health care use, either as a marker of better quality care, and/or as the mechanism by which needs identified during the assessment process might be addressed. Based on earlier reviews, and on the premise that earlier intervention would have larger effects, we also expected that assessments would be more effective in improving survival and physical functioning for women who began with better physical functioning at baseline compared to women who were already in poor health, and for those with lower mortality risk.

Materials and methods

Propensity score matching

For matching, participants who had an assessment were assigned into blocks based on the date of their first health assessment and corresponding to the interval between surveys (Fig 1). Propensity scores [23] were generated using logistic regression models and used to match participants who had never had an assessment with participants who had an assessment if they had a similar propensity score (within a 0.02 calliper distance matching algorithm), and if they had at least one unreferred visit in the same block. Matched pairs were retained only if the “no assessment” match had not died within three months of the health assessment date. Women who had an assessment but who could not be matched were excluded from the analysis.

Fig 1

Blocks and matched pairs for comparisons.

Unmatched women from the no assessment group were placed in the remaining subset for matching with women who had an assessment in the next block. This process was repeated for all five blocks. Covariate values for propensity scores were from the survey representing the start of the block. Models included 24 covariates (see Table 1), two squared terms, and between 8 to 66 interaction terms across the cohort blocks (See S1 Table). The c-statistic was used as a measure of propensity score performance through the model building process [24].

Table 1

Characteristics of women who have and have not had assessments, before and after matching.

	Characteristics at Survey 2			Characteristics after matching
	N = 9,099 women			N = 2,101 pairs (4202 women)
	No Health Assessment	Health Assessment	P-Value	No Health Assessment	Health Assessment	P-Value
	N = 2971 32.7%	N = 6128 67.3%		N = 2,101 50%	N = 2,101 50%
Marital Status
Not Widowed	55.8%	64.6%		60.2%	61.6%
Widowed	44.2%	35.4%	<0.001	39.8%	38.4%	0.975
Smoking Status
Never smoked	60.8%	63.7%		61.8%	61.9%
Ex/Current	39.3%	36.3%	0.009	38.2%	38.1%	0.799
Alcohol consumption*
Non-drinker			0.529			0.932
Low/rarely drinker	34.2%	33.1%		31.8%	31.4%
Risky/high risk drinker	62.2%	63.4%		64.5%	64.8%
Country of Birth*
Australian born			0.006			0.151
Other English speaking background	79.6%	76.5%		79.7%	76.7%
European	11.6%	14.2%		12.5%	14.9%
Other	7.1%	7.6%		7.1%	7.2%
Educational Qualifications
Up to Year 10	71.1%	71.4%		68.2%	68.2%
Post-school	28.9%	28.6%	0.154	31.8%	31.8%	1
Region
Major City	67.5%	65.5%		60.6%	60.6%
Regional/remote	32.6%	34.6%	0.057	39.4%	39.4%	0.801
Private Health insurance	46.9%	51.8%	<0.001	47.0%	45.9%	0.665
Manage on Income
Not too bad/Easy	76.4%	73.7%	0.007	76.1%	75.6%	0.916
General Health
Good/excellent	68.4%	74.4%		72.7%	72.0%
Fair/poor	31.6%	25.6%	<0.001	27.3%	28.0%	0.947
Visits to the GP
Up to four/year	40.7%	39.7%		46.6%	44.4%
More than four	59.3%	60.4%	0.35	53.4%	55.6%	0.732
Body Mass Index
Normal	52.3%	49.4%		53.3%	52.5%
Underweight	4.3%	2.9%		2.2%	2.6%
Overweight/Obese	43.5%	47.6%	<0.001	44.5%	44.9%	0.975
Co-morbidities
Diabetes	7.9%	7.0%	0.132	6.6%	6.5%	0.413
Heart disease	14.8%	12.4%	0.002	13.6%	13.6%	0.352
Asthma	13.3%	12.7%	0.456	12.5%	12.7%	0.891
Arthritis	40.6%	42.7%	0.059	41.1%	43.1%	0.925
Hypertension	33.1%	34.7%	0.13	34.4%	35.9%	0.592
Stroke	4.1%	2.4%	<0.001	3.3%	3.3%	1
Symptoms
Pain in the joints	43.8%	46.3%	0.029	43.8%	45.0%	0.877
Heart Symptoms	2.3%	2.3%	0.908	1.7%	2.9%	0.317
Urine Leakage	2.1%	1.5%	0.054	1.5%	1.9%	0.292
Feeling dizzy	3.8%	3.5%	0.373	3.5%	3.8%	0.621
Hearing Problems	13.6%	14.5%	0.26	11.0%	11.4%	0.962
Vision Problems	27.6%	25.0%	0.009	22.6%	22.9%	0.855
Falls	40.0%	36.7%	0.801	38.4%	39.2%	0.752
Need help with daily tasks*	9.6%	5.4%	<0.001	7.2%	6.5%	0.398
	Mean	Mean		Mean	Mean
	(Median)	(Median)		(Median)	(Median)
Age	75.4 (75.4)	75.3 (75.2)	0.007	76.53 (2.8)	76.5 (2.8)	0.976
Mental health Score	78.4 (84.0)	79.6 (84.0)	0.015	78.5 (16.0)	78.6 (16.0)	0.902
Physical Functioning Score	60.3 (65.0)	64.4 (70.0)	<0.001	56.6 (26.6)	56.6 (26.6)	0.996

* alcohol consumption, country of birth and need help with daily tasks were not included in propensity scores.

Results

A total of 9,099 women met all eligibility criteria and were classified as having had a health assessment (N = 6128, 67.3%) or never having a health assessment (N = 2,971, 32.7%). A total of 4,202 participants were matched using propensity score analysis.

Table 1 compares Survey 2 characteristics for women having assessments and no assessments for the whole sample and for the matched pairs. Prior to propensity score matching there were significant differences between participants who did and did not have an assessment. After propensity score matching, no statistically significant differences were evident between participants who did and did not have an assessment. The c-statistic values for the propensity models range around 0.5 indicating that each variable in the propensity score models is reasonably predictive of having an assessment. Variables country of birth, alcohol and needing help with daily living were not included in propensity models but were also balanced across the matched groups.

Women who had an assessment were less likely to have died over the follow-up period. Among all women, 62.7% of those who did not have a health assessment died, and 44.3% of those who did have an assessment died. Among the matched pairs, 66.2% of those who did not have a health assessment died, and 59.9% of those who did have an assessment died. However, 120 of the matched women died within three months of the health assessment date for the pair, and these 120 pairs were excluded. This left 3,204 women in block 1 (S2-23), 410 in block 2 (S3-S4), 156 in block 3 (S4-S5), 146 in block 4 (S5-S6) and 46 in block 5 (S6-S7) (Fig 1). The mean follow-up across all blocks was 8.5 years.

For Cox models assessing time to death, the proportional hazards assumption was violated for the health assessment variable, implying that the hazards varied with time since having the first health assessment. All covariates were found to be significant on univariable analysis and were entered into the multivariable Cox regression model using a stepwise process (Step 1). However, in the multivariable models, and once interactions with time (Step 2), and the interaction between assessment and time were added (Step 3), a number of covariates and interactions did not improve the model fit and these covariates were removed from further modelling.

Table 2 shows parameter estimates for Cox models for univariable analysis (Step 1), adjusting for significant interactions between covariates and time (Step 2), and finally the effects of assessments after all other significant variables have been included in the model.

Table 2

Predictors of time to death, based on stratified proportional hazards analysis.

Variable	Step 1	P-Value	Step 2: Treatment Effect	P-Value
	Parameter Estimate (Standard Error)		Parameter Estimate (Standard Error)
Age
75–80 years	Reference		Reference
>80–85 years	0.266 (0.051)	< .0001	0.266 (0.051)	< .0001
>85 years	0.448 (0.153)		0.411 (0.153)	0.007
GP
Less than four	Reference		Reference
More than four	0.143 (0.049)	0.004	0.147 (0.049)	0.003
Area of residence
Non-urban living	Reference		Reference
Urban living	-0.117 (0.046)	0.011	-0.118 (0.046)	0.01
CoMorbidities	0.086 (0.023)	<0.001	0.094 (0.024)	<0.001
Problems with Sight
No	Reference		Reference
Yes	0.072 (0.055)	0.19	0.083 (0.054)	0.127
Feeling Dizzy
No	Reference
Yes	0.08 (0.099)	0.419	-	-
Urinary Incontinence
No	Reference
Yes	0.122 (0.144)	0.394	-	-
Joint Pain and Symptoms
No	Reference		Reference
Yes	-0.102 (0.049)	0.038	-0.097 (0.049)	0.046
Physical function
Not Fair functioning	Reference		Reference
Fair functioning	0.348 (0.513)	0.498	0.331 (0.513)	0.518
Not Poor functioning	Reference		Reference
Poor functioning	1.674 (0.471)	<0.001	1.753 (0.469)	<0.002
Fair physical functioning by time
Fair functioning*log(t+1)	-0.036 (0.112)	0.749	-0.032 (0.111)	0.771
Poor physical functioning by time
Poor functioning*log(t+1)	-0.229 (0.103)	0.027	-0.244 (0.103)	0.018
Treatment (health assessment)	-	-	-3.508 (0.442)	<0.001
Treatment by time	-	-	0.693 (0.096)	<0.001
AIC	29857.018		29731.752

In these analyses, the hazards of death increased with age, more GP visits at baseline, multimorbidity, and poor physical functioning, and decreased for those having an assessment. The estimated hazard ratio associated with having a health assessment as a function of time t was given by: HR(t) = exp(-3.51 + 0.69t). Hazard ratios were 0.06, 0.24, and 0.99 at one, three and five years, respectively suggesting that assessment has an initial protective effect once time-by-treatment interactions was included. However, the parameter associated with having a health assessment and time interaction was positive, suggesting that the hazard ratios were increasing over time. This indicates that the protective effect of having a health assessment diminishes over time.

To further understand which women might have the most potential to benefit from assessments, and to account for heterogeneity in survival probabilities, we calculated prognostic scores based on the significant non-time dependent variables using the methods described by Putter et al. [25].

These prognostic scores were then used to divide the women into equal sized groups corresponding to low, medium and high risk of death. Women in the low risk group had a median survival time from baseline (Survey 2, 1999) of 178.3 months [IQR:116.9, 178.3], medium risk 147.5 months [IQR: 94.1, 178.3] and women of high risk had the shortest median survival time of 111.8 months [IQR:70.1, 163.4]. The mean age at baseline was 74.8 (SD:1.51) in the low risk group, 76.4 (SD: 2.07) in the medium risk group, and 78.4 (SD: 3.24) in the high risk group (overall mean 76.0, Std Dev: 2.4–3.5 years).

Women in the high risk group had a mean baseline physical functioning score (51.9, SD: 26.6) while women in the low and medium risk group had substantially higher physical functioning scores (67.9, SD: 23.8 and 61.3, SD: 25.8, respectively). Accordingly, there was a greater proportion of women in the high-risk group with poor physical functioning (37.2%) compared to the medium (24.3%) and low (16.8%) risk groups. Women from the high risk group also had a mean of 1.8 (95%CI:1.7–1.9) morbidities compared to women in the low and medium group (0.6 CI:0.6–0.6 vs 1.2 CI:1.1–1.2, respectively).

Among the medium and high-risk groups, a greater proportion of women lived in non-urban areas compared to women in the low risk groups (59.8%, 72.1% and 47.5%, respectively). Women in the high-risk group were more likely to have had more than four GP visits in the last 12 months. Given the women were matched, there were no differences in the proportions of women receiving health assessments across the three risk groups. However, women in the low risk group were more likely to be younger at the time of first assessment compared to women in the medium risk and high-risk groups. Approximately 343 (25.1%) women in the high-risk group were 80 years or older when they had their first health assessment (S2 Table).

Fig 2 shows the survival curves for women in the low mortality risk groups, their baseline physical functioning category (good, fair and poor), and whether they had an assessment or not. Approximately 70% of women grouped as low risk with good physical functioning who had a health assessment survived till the end of the study, compared to 59% of women with the same risk profile who did not have an assessment. Women in the low risk group who had fair or poor physical functioning had survival probabilities of 67% and 50%, respectively if they had an assessment and 55% and 34% respectively, if they did not have an assessment. Similar difference in survival probabilities were seen in the medium risk group (S1A Fig), although their overall survival time was less.

Fig 2

Model-based survival curves for women with low mortality risk.

Among women in the high-risk groups, marginal differences were seen between assessment and no assessment for women having good or fair physical functioning (S1B Fig).

Results of the multivariable logistic regression models of the association between health assessments and the composite outcome (poor physical functioning or death) at 1000 days and 2000 days follow-up are shown in Table 3.

Table 3

Predictors of poor physical functioning/death by 1000 and 2000 days.

Description	At 1000 days	At 2000 days
Physical functioning at the start of the block	OR [95%CI]	OR [95%CI]
Good functioning
Unreferred visit only	Reference	Reference
Health assessment	0.665, [0.519, 0.853]	0.964 [0.791, 1.175]
Fair functioning
Unreferred visit only	Reference	Reference
Health assessment	0.991, [0.776, 1.264]	0.665 [0.512, 0.865]
Poor functioning
Unreferred visit only	Reference	Reference
Health assessment	0.710, [0.501, 1.007]	0.494 [0.317, 0.77]
Perceived General Health
Fair/poor	Reference	Reference
Excellent/good/very good	0.529, [0.442, 0.634]	0.556 [0.455, 0.681]
Area of Residence
Non-Urban	Reference	Reference
Urban	0.828, [0.706, 0.971]	0.932 [0.801, 1.085]
Age in years	1.064, [1.008, 1.124]	1.13 [1.073, 1.19]

At 1000 days: Adjusted for Urinary Incontinence.

At 2000 days: Adjusted for BMI, number of comorbidities, problems with sight, problems with hearing, falls in the past 12 months.

There was a signficant interaction between physical functioning assessed at the start of the relevant block, having a health assessment, and poor outcome. Participants who had good baseline physcial functioning scores and who had an assessment had significantly lower odds of poor outcomes at 1000 days follow-up compared to participants having an unreferred visit only (OR: 0.67, 95%CI: 0.52, 0.85). There were no differences between assessment and no assessment groups for participants having fair or poor physcial functioning at baseline. Other factors that were protective against a poor outcome included living in a major metropolitan area (OR: 0.83 95%CI: 0.71, 0.97) and having excellent, very good or good self-rated health (OR: 0.529, 95%CI: 0.442, 0.634). Each one year increase in age resulted in a 6% increase in odds of poorer outcomes at 1000 days follow-up.

At 2000 days from assessent, participants with fair functioning at baseline who had a health assessment had almost 40% reduced odds of poor health outcome compared to women who only had an unreferred visit (OR:0.67, 95%CI:0.51, 0.87). Similar results were seen for those with poor physical (OR:0.49, 95%CI: 0.32, 0.77). Higher odds of poor outcomes at 2000 days were also observed for women who were overweight (OR: 1.59, 95%CI: 1.01, 2.50), and those with two or more conditions (OR: 1.17, 95%CI:1.01, 1.37).

The mean difference in health service utilisation between participants who had an assessment and those who had an unreferred visit only is shown in Fig 3 (Blocks 1, 2 and 3) and S2 Fig (Blocks 4 and 5). Regardless of the cohort block, participants who had an assessment had significantly higher use of health services both before and after the assessments, and a further increase in the three years following the assessments.

Fig 3

Health service utilisation for women who had an assessment and those who did not (Blocks 1,2,3).

Discussion

This large observational study used propensity score matching to compare groups of women who had and had not had assessments in terms of survival, physical functioning, and health service use.

Women who had health assessments had better survival and were less likely to have the composite outcome of poor health or death. These findings add to the evidence from numerous large randomised controlled trials to show that health assessments can improve health outcomes. Beyond these previous trials, the study also shows how assessments are used at a population level and over time, as well as their potential effects on health outcomes. Unlike earlier studies where assessments were conducted within the context of a trial with protocols for the administration of the assessment and follow-up [12], this study evaluates assessments according to their uptake by women and to how they are delivered by general practitioners within the health system. While this observational approach provides evidence as to how assessments perform in practice, there are limitations in terms of the study rigour compared to an experimental trial design including the potential for incomplete matching of assessed and non-assessed groups (discussed later). The external validity of both trial and observational approaches can be limited, with the former affected by participation bias, and the latter potentially limited by matching, as seen in this study. In this study the matched pairs were on average one year older than the overall group of women who had assessments, they also had fewer GP visits and lower physical functioning scores.

The differences in survival may be explained by the higher overall health care use by the women who had assessments, both before and after the assessment. It may be that women who have assessments are those who are in more frequent contact with the health care system [14], and receive better quality care. We did observe an increase in health care use following the assessments suggesting also that more intensive health care use may be a mechanism for improved health outcomes among assessed women. This is consistent with a previous review that concluded that health assessment programs were particularly effective if they incorporated ongoing follow-up [3]. In their review, Beswick et al. [1] discuss the discrepancy between their findings and the results of a large MRC trial which found that assessments were not effective. The authors of the MRC trial note that assessments were not generally well conducted, and this may have limited their effectiveness [26].

Beswick et al. [1] also consider the effectiveness of the health system overall as an explanation for this discrepancy between the results of controlled studies and the MRC trial.

More recent research has shown some additional evidence for the potential for health assessments to reduce mortality rates. A randomised controlled trial in Germany showed a 20% reduction in mortality rates for those who had a health assessment [27]. Also, a recent retrospective analysis using linked MBS and aged care data to assess the use and impacts of health assessments for people using aged care, showed that after propensity score matching, people who had an assessment had 7% lower risk of death with a mean follow-up of 1.9 years. However, those who received assessments had a higher risk of admission to permanent residential aged care [28].

In our study, the logistic regression analysis showed that those with good physical functioning were less likely to have poor outcomes at 1000 days if they had an assessment. This is consistent with the review by Beswick et al. [1], which found more evidence for the effectiveness of assessments for older people in the general community than for assessments targeted to frail older people. However, few trials follow people for as long as 2000 days. At 2000 days, we found that there were significantly lower odds of poor health outcome for those with fair or poor physical functioning if they had an assessment. Assessments may have a particular role for improving outcomes for those with poor physical functioning but long survival probabilities. This possibility is supported by our analysis showing that where women had low or medium mortality risk, there were larger survival differences between women who had an assessment and those who did not, even if they had poor physical functioning.

That assessments on their own would have such long lasting effects in frail older people is unlikely, but it may be that they set people on a better trajectory of health service use.

It should however also be noted that people with higher physical functioning scores had better chances of survival and these survival probabilities were higher if people had an assessment.

In undertaking these analyses we have been extremely careful to avoid immortality bias, whereby the women have to survive long enough to have assessments (and therefore have an artefactual survival advantage), using propensity matching, removing matches who died within three months of the assessment date, and using prognostic stratification.

Propensity matching also increases the comparability of the groups in terms of other factors which may affect survival or quality of life. Propensity methods are increasingly used in clinical research to enable comparisons of treated and untreated populations using observational data [29]. Our approach using 1:1 matching, has an advantage of producing balanced groups for known confounders, however there is a risk that some residual confounding may still bias the results, and propensity matching cannot account for unknown confounders in the same way that a randomised trial can. A disadvantage of 1:1 matching is that the resultant matched pairs may differ from the original population.

However, for women having health assessments, the differences between matched an unmatched samples in our study are small, due to the large pools available for matching. The matched sample were slightly older and better educated, had poorer median physical functioning score, had fewer GP visits, and included fewer non-smokers, more people in non-metropolitan areas, and fewer people with private health insurance.

Other strengths of this study include the long-term follow-up and complete outcomes for a large cohort of women in terms of health service use and death.

Ascertainment of physical functioning outcomes was less complete as this was dependent on participation in the survey. Attrition rates for ALSWH are around 10% for each survey and have been reported elsewhere [30–33].

Conclusion

Health assessments have potential to improve survival and physical functioning of older people, but their uptake and their effect appears to be strongly coincident with greater use of health services. If assessments are effective in improving health care use and outcomes, then attempts to extend assessments to reach underserved women may increase their overall effectiveness.

The value of assessments may be strongly linked to the quality and effectiveness of health services overall. Assessments may be a key gateway to better care, and it is important to understand the key elements of effective care for older people and how they may be improved through the assessment process.