Chapter 3 - Model Construction: A model of an Ageing HIV Population
3.4 Model Checks
3.4.5. Co-medication
A final model check that was carried out was to compare the model output on CV medication to observational data from the ATHENA cohort. The results of this comparison are shown in Table 3.17 and show that the model manages to capture CVD co-medication well. The exception is an underestimation in the number of patients starting ACE inhibitors, statins, and diuretics in 2011.
Chapter 3 – Model construction
Table 3.16. The annual number of new co-morbidities developed by HIV-patients according to the observational data from the ATHENA cohort and the model output.
2011 2012
Data Model Data Model
Diabetes 63 74 54 83
Hypertension 418 438 333 477
Hypercholesterol 470 598 658 655
Malignancy 75 69 71 72
MI 20 32 19 42
Osteoporosis 238 111 94 96
Renal insufficiency 63 76 54 87
Stroke 18 30 13 31
Table 3.17. The number of HIV-patients who start a co-medication according to the observational data from the ATHENA cohort and the model output.
2011 2012
Data Model Data Model
Ace inhibitor 224 132 184 150
Beta blockers 186 138 146 144
Calcium blockers 66 52 61 46
Diuretics 309 137 208 146
Statins 361 208 298 255
3.5 Discussion
This chapter provides a detailed description of the construction of an individual-based model of the ageing HIV-population in the Netherlands. The model follows HIV-patients from the start of treatment, as they age, develop co-morbidities, namely diabetes, hypertension, hypercholesterol, osteoporosis and renal insufficiency, experience strokes, MIs or malignancies and are prescribed co-administered treatment for these conditions. This individual-based model provides a framework with which to quantify the future challenges posed by an ageing HIV-population, by predicting trends in
Chapter 3 – Model construction
The model captures the key aspects affecting clinical care of HIV-patients, including the major age-related co-morbidities, their common physiological pathways and their treatment. The design of the model was informed by consultations with treating physicians in the Netherlands, in order to incorporate the main clinical aspects involved in the care of these patients. The parameterisation of the model was accomplished using data from the national ATHENA cohort, a non-selective dataset of all HIV-patients in clinical care in the Netherlands. Where the data were insufficient, in depth literature reviews were carried out, selecting, where possible, data from large international studies. The incorporation of these common causal pathways allows the model to capture the natural aggregation of morbid disease within patients (that patients with a certain co-morbidity are at an increased risk of another co-co-morbidity through common causal pathways), rather than simply model the development of co-morbid disease as a function of age and sex.
Numerous model checks were completed in order to validate certain assumptions, for example the linear increase in mean age. All checks showed that the parameterization of the model concurred with ATHENA data, at least over the short-term. The deterministic model of HIV-infection that was incorporated into the individual model, to predict the number of people starting treatment in the future, is shown to be an adequate method for predicting future trends in HIV-infection in the Netherlands. In the future, this could be updated with new incidence estimates from an in-depth incidence model of the Dutch epidemic by Dr van Sighem (unpublished). The synthesis of the demographic and clinical factors affecting the development of co-morbidities in the model has output data similar to those observed in the ATHENA cohort, and modelling of co-medication appeared to be in line with observational data. Finally, checks on the model’s output of mortality showed that the way the mortality data reported by the D:A:D Study Group was used to simulate deaths was a valid approach. Mortality amongst HIV-patients was higher than in the general population, with SMRs within the 95% CI reported by larger European HIV cohorts, and with annual percentage of deaths amongst treated patients similar to those reported by the ATHENA cohort.
Despite the strengths of the model’s design and parameterization, it has a number of limitations. The model simulates major age-related co-morbidities, namely CVD (hypertension, hypercholesterol, MI and stroke), diabetes, osteoporosis, malignancies, and renal insufficiency. The ATHENA cohort collected these age-related co-morbidities; however physicians do not record them systematically. This and the fact that not all age-related co-morbidities and their treatment are modelled means that trends produced by this model will underestimate the true future burden of co-morbid disease and co-medication, as well as the consequent impact on drug interactions and long-term ARV restrictions. In addition, the model does not currently simulate recurrent MIs, strokes or malignancies. In the future, the model should be further developed to include recurrent CV events by using data on recurrence rates and to incorporate a CVD framework, like the Framingham or SCORE framework, to better capture CVD.
Chapter 3 – Model construction
In the future it may be valuable to model other age-related co-morbidities, such as psychiatric illnesses and neurodegenerative impairments. Neurocognitive conditions are a major problem amongst older HIV-infected patients, with studies showing that older HIV-patients are more likely to develop neurocognitive problems such as dementia than the age-matched uninfected population [15,309–312]. Some neurocognitive, age-related co-morbidities such as late-onset Alzheimer and late onset Parkinson have a mean age of onset of around 74 and 61 years, respectively [313,314]. So while mean age of onset of Parkinson’s and Alzheimer do not currently make them major age-related co-morbidities amongst HIV-patients, they will become increasingly important in the future as this population ages. Additional data on these co-morbidities in HIV-patients will be important in the future. It will allow their incorporation into the model and add to the growing picture of the future challenges posed by the ageing HIV-population.
Conditions related to memory loss will become particularly important for the treatment efficacy of multi-morbid HIV-patients. In HIV-patients, dementia can be the result of a number of causes, including late-onset Alzheimer, HIV-associated mild neurocognitive disorder or AIDS-related dementia [15,309–312]. Studies have shown that HIV can enter the central nervous system via infected macrophages. This can trigger a cascade of events, including the release of neurotoxins that affect the formation of memory [315]. Loss of memory can complicate the treatment of HIV-patients.
Patients can find it difficult to adhere to their treatment regimens, forgetting pills, particularly in light of the ever increasing pill burden, thus compromising the effectiveness of both HIV and co-administered treatments.
Future expansions of the model may also benefit from modelling individual cancers or groups of cancers, rather than non-AIDS malignancies as a whole. There are many types of cancers, all of which differ starkly in their risk factors, treatment and long-term outcomes [316]. It is their large number and incidence rate that make them difficult to study in more detail in HIV-infected people.
For example, the incidence of breast cancer in the UK (the most common cancer in the UK), is 157 new cases per 100,000 women [317]. The number of HIV-patients in the Netherlands does not constitute a large enough population to study the large number of different cancers in any greater detail. Large multi-cohort studies, like the D:A:D Study into adverse events, may provide more data in the future [318]. Interestingly, although HIV-patients are at an increased risk of cancer compared to the general population, age of onset may only be minimally younger compared to the general population [319]. Furthermore, some related cancers have not been found to be increased in
age-Chapter 3 – Model construction
other co-morbidities with common causal pathways. Subclinical hypertension and hypercholesterol have been shown to be risk factors for other CVD. A study by Kuller and colleagues showed that patients with subclinical cardiovascular disease were at higher risk of total coronary heart disease including CVD-related death and non-fatal MIs than those without subclinical disease [320].
The physiological mechanisms that affect the development of co-morbidities are numerous and multi-factorial. There is a large body of literature on factors affecting the development of different co-morbidities, including lifestyle factors. CVD, for example, has been associated with a large number of different risk factors, including age, sex, hypertension, diabetes mellitus, smoking, dyslipidaemia (abnormal amounts of lipids in the blood) [321–324], and smoking is a well-established risk factor for lung cancer [234]. In fact, lifestyle factors can be a more important risk factor than HIV-infection or HIV-treatment [232,233]. There were insufficient data on lifestyle in the ATHENA cohort to incorporate it into the model. In the future patients could be divided into ‘healthy’ and
‘unhealthy’ lifestyle category, contributing in this way to their risk of developing co-morbid diseases such as CVD, and cancers. Another approach would be to include a risk framework, such as the Framingham or D:A:D framework to better predict the development of co-morbidities.
ARV use has also been associated with the development of a number of co-morbidities. HIV-treatment can cause toxicities affects virtually all organ systems. Reported toxicities include diabetes, hepatotoxicity, renal insufficiency, dyslipidaemia, pancreatitis, neuropathy and lactic acidosis [89,218–223] with NNRTIs, Abavcavir and some PIs, for example, associated with cardiovascular toxicity [224,225,325] and PIs with osteoporosis [209]. The impact of ARV on patients is complex, with some ARVs changing the bioavailability or metabolism of other ARVs. The decision was made not include the effect of ARVs on co-morbidities into the model at this stage. The historic impact of ARV on the development of co-morbidities cannot be disentangled entirely from the current work. It is likely that the model parameters used in this model reflect some aspects of past ARV use, like the impact of ABC on CVD.
Mechanisms that link co-morbidities may also have a time-dependent aspect, with risks for other co-morbidities being cumulative in the long-term. For example, a patient who has hypertension is at an increased risk of other cardiovascular events, such as an MI [321–324]. However, the patient does not have the same increased risk of experiencing an MI the day his blood pressure rises to above the clinical threshold for hypertension. Rather, his hypertension is going to cause long-term damage to his or her cardiovascular system which in the longer-term will put him at an increased risk of an MI. At this stage the model does not capture this aspect. However, the fact that co-morbidities are modelled as a function of age may capture some aspects of this time-dependence in the model.
One potential source of information that could address many of these limitations is the AGEhIV Cohort Study. The AGEhIV Cohort Study is a prospective cohort study in the Netherlands that was established in 2010. Its aims are to investigate the prevalence and incidence of a broad range of age-related co-morbidities and their risk factors in HIV-infected patients and non-HIV-infected
Chapter 3 – Model construction
controls [326]. It collects a number of clinical markers with the goal of providing data on the independent effect of HIV-infection on the development of a number of co-morbidities. The study should allow the model to be expanded, by including other age-related co-morbidities such as Alzheimer’s, Parkinson’s and other conditions. It will also be able to provide parameters on risk factors for these co-morbidities, including lifestyle factors. These results can be expected over the next few years.
One aspect of clinical care that is difficult to capture in a model is the complex and multi-factorial nature of prescribing medication in multi-morbid HIV-patients, including prescription practices and behaviour of clinicians. Balancing harm and benefit to the patients as well as cost to the health care system is paramount [191]. By designing the model with the consultation of treating physicians, it was possible to capture some of the main aspects of clinical care of HIV-patients, such as the exclusion of co-medication like simvastatin that is contra-indicated in HIV-patients [60,327] . However, the treatment of multi-morbid HIV-patients is a complex area of clinical care. Patients differ greatly in terms of their individual medical needs and risk factors. Two patients who present with hyperglycaemia may be treated very differently by the same physician. Their treatment will depend on a number of factors, including their level of hyperglycaemia, age, family history, and medical history [60]. An elderly patient with hypertension and diabetes may be put on preventative treatment for more serious CVD as well as treatment for their diabetes. A younger patient with no other co-morbidities, who presents with hyperglycaemia, may be encouraged to make lifestyle changes prior to putting him or her on treatment. These individual factors mean that in reality, treatment of patients can be very complex.
The model has to make assumptions about the co-administration of medication. For example, the model assumes that the co-administration of drugs is independent of age and sex. The model further assumes that a patient’s need for co-medication is only evaluated after they are diagnosed with a co-morbidity. The model does not allow for patients to have this need revaluated over time. While this is not what occurs in reality, it is in line with the aims of this model; that is, to quantify the burden of polypharmacy and its effects on HIV-treatment at one point in time.
In the model, it is assumed that prescribing behaviour will continue to follow a similar pattern to that observed over the last 15 years. Prescribing behaviour may change in response to an ageing population in that whereas only a proportion of co-morbid patients currently receiving co-medication, this may change with ageing of patients in clinical care. As mentioned, the clinical care of patients
Chapter 3 – Model construction
The model could have been designed with increased complexity. For example, it could have included assumptions surrounding lifestyle factors or modelled additional co-morbidities, as described above. However, this would have come at the cost of the reliability of the output. The fact that the model was designed after consulting treating physicians was created using a large non-selective database that collects data from all HIV-infected patients in clinical care in the Netherlands, and is informed by an extensive literature review means this is a very powerful model. The right balance was achieved, utilizing data, literature and assumptions to create a robust model. To my knowledge this is the first model to capture key aspects of an ageing Dutch HIV-population, including co-morbidity, polypharmacy and cART and it will be valuable tool in providing a picture of the future challenges in clinical care of an ageing HIV-population in the Netherlands. The future aim is to further expand this model through continued collaboration with the ATHENA cohort and the AGEhIV Cohort Study and to adapt the model to new settings through new collaborations. The direction of future work is outlined in detail in Chapter 6 (Section 6.5 Future Work, page 159), and includes the numerous points outlined above.
In summary, an individual-based model of an ageing HIV-population in the Netherlands was constructed. This model can be used to predict the future age-structure, and the burden of age-related co-morbidity and polypharmacy on HIV care, as well as to the impact of interventions aimed at improving patient outcome and clinical care. It provides a powerful tool that can be used in a range of applications, and can be adapted to a number of different settings to quantify the future demands on the health care system as well as to inform policy on how to ensure a continued high standard of care.