- Email: [email protected]
Prostate cancer: a systems approach overview
Mini-symposium: Pathology of the Prostate
Prostate cancer: a systems approach overview
more difficult and has less resolution than direct inspection. This leads to fewer biopsies and less participation of histopathologists in the whole spectrum of a disease. This restriction of biopsy material to a few points along the disease spectrum can lead to a distortion of histopathologists’ perception of the disease and some inflexibility when faced with changes in management of that disease, e.g. introduction of a cancer screening programme. A systems approach, often called ‘systems thinking’, is a methodology that sets out to create an overall model of a particular situation, which can be used to make predictions about that situation if specific elements within it are changed. The systems approach is often divided into two categories – soft and hard systems. A soft systems approach identifies all the elements within a system and the qualitative relationships between these. A hard systems approach goes further, taking quantitative data and producing a model that will produce quantitative predictions about outcomes from changes in input data. A hard systems approach will be more familiar to scientists since much translational molecular pathology research contains an element of hard systems modelling, e.g. development of a novel molecular marker that will predict systemic metastases in node negative breast cancer. The soft systems approach has had less application in the scientific and biomedical arenas and is often derided in those areas because it lacks a quantitative basis. However, soft systems methodologies have had widespread application in management ‘science’ where they have been shown to be very successful at addressing complex problems.1–4 Since diagnostic histopathology is often a blend of scientific knowledge integrated within a management context, it is likely that a soft systems approach could yield valuable insights into the effects of change in the scientific knowledge or the management of diseases. This review applies some systems approaches to prostate cancer and shows how these can focus our thoughts on the scientific facts about the disease as well as enabling us to predict how changes in management of the disease will affect histopathology services. The review starts with overly simplistic models but then builds up to more complex models that should have some clinical utility.
Simon S Cross Freddie C Hamdy John R Goepel Robert F Harrison
Abstract Systems thinking is a set of methodologies that facilitate analysis and predictions for a complex system in a holistic way. Prostate cancer has precursor stages that are difficult to detect and a number of different therapy options for different stages, so it is a complex disease to manage within healthcare systems. In this review we show how systems thinking, especially causal loop diagrams, can be a very valuable tool for gaining greater insight into the pathogenesis and diagnosis of prostate cancer and to predict the consequences in major changes in the pattern of healthcare for this disease, e.g. introduction of national screening programmes using serum prostate specific antigen. The systems thinking approach can be used to predict changes in histopathology workflow when other parts of the system are changed.
Keywords causal loop diagrams; influence diagrams; prostate cancer; prostate cancer screening; prostatic intra-epithelial neoplasia; prostate specific antigen; state diagrams; systems thinking
Introduction Histopathology has an intermittent relationship with the management of many diseases because the opportunities for biopsy and histopathological examination are often limited to a few points along the disease progression. Some organs, the colorectum being an exemplar, are relatively accessible for visual inspection and biopsy so a disease such as colorectal cancer is well-represented in histopathology biopsies from early precursor lesions through to advanced disease. Other organs, including the prostate, are not available for visual inspection and the imaging and biopsy of these organs is
Phase I modelling – a simple state diagram The simplest level of a systems approach is to identify the elements within a system without ordering them or defining any relationships between them. This may appear completely selfevident, and when one has previous knowledge of the subject it may appear unnecessary, but it is important to do this with as much inclusion as possible so that no potentially important elements are left out of the system. It becomes extremely important to do this when trying to model a new system about which there is very little initial information. If we think about prostate cancer at its most basic level we know that men are born with (we assume) normal prostates but that a sizeable proportion of men develop prostate cancer so at its very simplest level we can identify two states – normal and prostate cancer (Figure 1).
Simon S Cross MD FRCPath is at the Academic Unit of Pathology, School of Medicine & Biomedical Sciences, University of Sheffield, UK. Freddie C Hamdy FRCS is at the Academic Unit of Urology, School of Medicine & Biomedical Sciences, University of Sheffield, UK. John R Goepel FRCPath is at the Department of Histopathology, Royal Hallamshire Hospital, Sheffield Teaching Hospitals NHS Foundation Trust, Sheffield UK.
Normal prostate
Robert F Harrison Department of Automatic Control & Systems Engineering, University of Sheffield, UK.
DIAGNOSTIC HISTOPATHOLOGY 14:3
Prostate cancer
Figure 1 A simple two state representation of normality and prostate cancer.
122
© 2008 Elsevier Ltd. All rights reserved.
Mini-symposium: Pathology of the Prostate
Of course we know that there are many more entities in the prostate cancer spectrum and so we can define those to give more elements in our state diagram (Figure 2). This state diagram has no order to the elements, they are simply put onto the page in the order that we have thought of them, which would seem to be unhelpful but it is worth reflecting that histopathology when performed in an uncritical manner would be somewhat analogous to such a diagram. In histopathology we are taught to identify reliably discrete morphological patterns and to assign them agreed labels, which are included in our reports and should be understood by clinicians reading those reports. Diagnostic histopathology can run at this most basic level without much detriment to the overall management of patients as long as the labels can be applied to the morphological patterns with a one-to-one correspondence and a high level of reliability. However, it is much better if the elements are more ordered with clear definitions of their relationships because this gives a much greater understanding of the interaction between diagnostic histopathology and overall patient management, especially in difficult areas, e.g. atypical small acinar proliferation and early prostate cancer in needle core biopsies. Looking at the simple state diagram in Figure 2 there are still some important points that arise. Among the entities included are neuroendocrine prostate cancer and large duct prostate cancer (shown in red in Figure 3). When we consider those in the context of the other elements we realise that, although they are variants of prostate cancer, they are not going to fit in with the overall progression that we are going to construct. This will be based on the most common adenocarcinoma of the prostate that arises from the acini in the prostate with no special pattern of differentiation. These other entities would need to be included in a complete systems model of prostate cancer but are likely to add unnecessary complexity to our initial model so we will omit them. It also means that we need to tighten up on our terminology for the existing elements so we end the first phase of modelling with the state diagram in Figure 4.
Phase II modelling – an influence/progression diagram Now that we have defined elements that we wish to include in the model, we can look at the relationships between these elements and indicate them on the diagram by arrows. When we look at the later stages of prostate cancer the elements are relatively easy to order. Localised prostate cancer can progress to locally advanced prostate cancer, which could, in turn, progress to metastatic prostate cancer as shown in Figure 5. Many diagrams like this have been reproduced in the pathology literature, Walter Bodmer’s progression from colorectal adenoma through to invasive colorectal carcinoma being the prototypical example. It is important to get all the details of such sequences correct as such diagrams spread rapidly through the literature, presumably owing to the much easier assimilation of information from diagrams rather than blocks of dense text. Looking at our new diagram (Figure 5) we need to look at all the elements and ensure that all appropriate arrows have been included to indicate all possible relationships. We ask ourselves whether prostate cancer can metastasise from localised, as well as locally advanced, prostate cancer. That can indeed occur so we need to draw an arrow from localised to metastatic prostate cancer, which does not pass through locally advanced prostate cancer as an obligate transitional stage (Figure 6). We can also look at whether the arrows should be one way or two way, i.e. can locally advanced prostate cancer regress to localised or metastatic prostate cancer regress to localised prostate cancer? The answers to these questions are obviously no, at least without some very effective therapy, so the arrows should be left one way. Including the precursor lesions to invasive prostate cancer in the influence diagram is more difficult and this is where the value of these diagrams becomes evident. Prostatic intra-epithelial neoplasia (PIN) is probably the easier of the two to deal with. What we need to know is whether PIN is a precursor from which invasive prostate cancer develops and if so is it an obligate precursor,
Neuroendocrine prostate cancer
Low grade prostatic intraepithelial neoplasia (PIN)
Normal prostate
Atypical small acinar proliferation
Locally advanced prostate cancer
Localised prostate cancer
Metastatic prostate cancer
Large duct prostate cancer
High grade prostatic intraepithelial neoplasia (PIN) Figure 2 A state diagram showing all the elements that might be considered for a model of prostate cancer pathogenesis and progression. The elements are not ordered in any way.
DIAGNOSTIC HISTOPATHOLOGY 14:3
123
© 2008 Elsevier Ltd. All rights reserved.
Mini-symposium: Pathology of the Prostate
Neuroendocrine prostate cancer
Low grade prostatic intraepithelial neoplasia (PIN)
Normal prostate
Large duct prostate cancer
Metastatic prostate cancer
Atypical small acinar proliferation
Locally advanced prostate cancer
Localised prostate cancer
High grade prostatic intraepithelial neoplasia (PIN) Figure 3 The state diagram from Figure 2 has been taken and two elements, neuroendocrine prostate cancer and large duct cancer, have been highlighted in red because they do not fit into a general system of the most common non-specialised prostate cancer.
i.e. do all cases of ordinary prostatic adenocarcinoma arise from foci of PIN. If PIN is not a precursor but only a marker of risk of development of prostate cancer then it will not be included in the direct disease progression sequence. If it is a precursor but not an obligate one, then it will be in the sequence but there will be an alternative parallel route. These options are illustrated in Figure 7. The data to decide between these different options is quite difficult to acquire for the prostate because of the previously discussed
problems with imaging and taking targeted biopsies from the organ. In the colorectum the picture is much clearer because abnormalities can be seen at an almost unicryptal level as aberrant crypt foci using magnifying chromoendoscopy and can even be subtyped into hyperplastic and adenomatous aberrant crypt foci. Thus, the current theory in colorectal cancer is that aberrant crypt foci are obligate precursors of colorectal cancer. In the prostate such small precursor lesions cannot be identified in vivo and the main source
Low grade prostatic intraepithelial neoplasia (PIN)
Normal prostatic acinus
Atypical small acinar proliferation
Localised prostatic adenocarcinoma
Locally advanced prostatic adenocarcinoma
Metastatic prostatic adenocarcinoma
High grade prostatic intraepithelial neoplasia (PIN) Figure 4 The state diagram from Figure 3 has been taken, the two highlighted elements have been omitted and the terminology in the other elements has been altered to be more specific, e.g. prostatic adenocarcinoma rather than prostate cancer.
DIAGNOSTIC HISTOPATHOLOGY 14:3
124
© 2008 Elsevier Ltd. All rights reserved.
Mini-symposium: Pathology of the Prostate
Localised prostatic adenocarcinoma
Locally advanced prostatic adenocarcinoma
Metastatic prostatic adenocarcinoma
Figure 5 The elements in the later stages of prostate cancer have been ordered into a sequence.
suboptimal histological processing.8 It is, therefore, used as a histological term that suggests rebiopsy in order to obtain sufficient histological material to make a definitive diagnosis of invasive prostate cancer and is thus not a biological precursor of prostate cancer. One piece of evidence that provides scant evidence that it could be a precursor, or at least a true biological marker of risk, is that if prostates are rebiopsied following a diagnosis of atypical small acinar proliferation in unilateral biopsies, then in 39% of cases the invasive cancer was found exclusively in a site distant from the initial biopsies.8 This is confusing since if atypical small acinar proliferation was simply a term relating to invasive cancer that could not be diagnosed in the particular biopsy under consideration, then one would expect cancers diagnosed immediately after the first sample to be at the same site as the initial biopsies. However, on current published evidence and definitions, atypical small acinar proliferation cannot be included as a bona fide biological precursor of invasive prostate cancer. Figure 8 now looks like a reasonable summary of the progression of prostate cancer but we need to include one more factor in the later stages of the disease – androgen dependency.9 This is a factor that we rarely encounter in histopathology because we do not test prostate cancers for androgen receptors in the same way that we do breast cancers for oestrogen and progesterone receptors. It is, however, a vital factor in determining treatment of non-localised prostate cancer and so needs to be included in our system. We have included this in Figure 9 on the basis that, when it arises, prostate cancer will be androgen sensitive but at any point after that it may become androgen insensitive whether by spontaneous mutations within the tumour in an untreated patient or by selection of androgen-insensitive clones of the tumour in patients receiving androgen ablation therapy. We also know that androgen-insensitive tumours do not become androgen dependent again by further accumulated mutations, so the arrows are uni-directional.10
of information is from painstaking examination of radical prostatectomy specimens where the main identified invasive prostatic adenocarcinoma is likely to have outgrown and destroyed any precursor lesions. The current published literature strongly suggests that PIN is a direct precursor of prostate cancer.5–7 It is found in the same areas of the prostate as invasive cancer with the same relative frequency as prostate cancer in different groups, i.e. increasing incidence with increasing age and increased incidence in groups with known increased incidence of prostate cancer. Many of the molecular abnormalities of prostate cancer are also found in PIN but not in the morphologically normal background acini in prostates that contain cancer. No conclusive evidence appears to exist to suggest that all prostate cancer arises from PIN; this may be due to the problems with detection but means that, at present, an alternative parallel pathway has to be included. In Figure 4 we split PIN into low and high grade because these are morphological categories used in diagnostic histopathology. We know that we only include high grade PIN in our histopathology reports because the intra- and interobserver variability for this diagnosis is within reasonable limits whereas the same parameters for low grade PIN are not sufficiently reliable to act as a basis for any clinical decisions. However, if we are looking at PIN in the context of the spectrum of prostate cancer, it is more logical to include it as a single category rather than two rather arbitrary morphological categories. The final feature that we need to consider is whether PIN can regress, i.e. whether the arrow between normal prostatic acini and PIN should be one or two way? There is considerable evidence in the literature that PIN can regress under treatment conditions, such as androgen ablation therapy, and some evidence that lesser changes in the prostatic milieu, such as selenium or vitamin E as dietary supplements, may also induce regression – so a two-way arrow or separate arrows in opposite directions seems reasonable (Figure 8).5–7 The place of atypical small acinar proliferation as a precursor of prostate cancer is less secure. The published definitions of this term imply that it is a histological appearance that is suspicious of invasive prostatic adenocarcinoma but that the definitive diagnosis cannot be made because of some compromising features of the biopsy such as the small size of the focus of abnormality, some abnormal features of morphology, e.g. atrophy, or artefacts from
Localised prostatic adenocarcinoma
Phase III modelling – using the influence diagram Now that we have constructed a fairly comprehensive influence diagram for the whole spectrum of prostate cancer, we can use this
Locally advanced prostatic adenocarcinoma
Metastatic prostatic adenocarcinoma
Figure 6 An additional loop has been added to the diagram from Figure 5 because it is recognised that localised prostatic adenocarcinoma can metastasise without necessarily passing through a locally advanced stage.
DIAGNOSTIC HISTOPATHOLOGY 14:3
125
© 2008 Elsevier Ltd. All rights reserved.
Mini-symposium: Pathology of the Prostate
a
PIN is an obligate precursor of prostatic adenocarcinoma
Normal prostatic acinus
b
Localised prostatic adenocarcinoma
PIN is a non-obligate precursor of prostatic adenocarcinoma
Normal prostatic acinus
c
Prostatic intraepithelial neoplasia (PIN)
Prostatic intraepithelial neoplasia (PIN)
Localised prostatic adenocarcinoma
PIN is not a precursor of prostatic adenocarcinoma
Localised prostatic adenocarcinoma
Normal prostatic acinus
Prostatic intraepithelial neoplasia (PIN) Figure 7 Three different possible roles of prostatic intra-epithelial neoplasia (PIN) in the pathogenesis of prostate cancer are shown. a PIN is an obligate precursor of b PIN is a non-obligate precursor of prostatic adenocarcinoma, i.e. some but not all cases of prostatic adenocarcinoma rise from PIN. c PIN is not a precursor of prostatic adenocarcinoma, i.e. PIN is a marker of increased risk of development of prostatic adenocarcinoma but it does not arise from it. Note how the diagrams summarise these possibilities much more succinctly than text.
in our thinking about the disease. It is easy and instructive to plot an individual patient’s pathway through the system to the point at which they present to the healthcare system. Figure 10 shows the pathway of a man who presented to his general practitioner with back pain and was found to have sclerotic bone metastases in his lumbar spine and a raised serum prostate specific antigen (PSA). What we can see immediately (and which would have to be determined from some clinical information, such as response to androgen ablation therapy) is that, although metastatic, his
Normal prostatic acinus
Prostatic intraepithelial neoplasia (PIN)
tumour is androgen dependent so he should have a response to androgen ablation therapy. Since he is clearly presenting at a late stage of disease and will have been through many earlier stages, then it is instructive to see whether any healthcare interventions made earlier would have prevented such a late stage presentation. If he had no symptoms of prostatism (and he need not have since most prostate cancers are relatively small and peripheral) then he would have had no cause to present to the healthcare system until the onset of his back pain. This means that only some form
Localised prostatic adenocarcinoma
Locally advanced prostatic adenocarcinoma
Metastatic prostatic adenocarcinoma
Figure 8 The complete sequence of prostate cancer pathogenesis and progression from normal prostate through to metastatic disease.
DIAGNOSTIC HISTOPATHOLOGY 14:3
126
© 2008 Elsevier Ltd. All rights reserved.
Mini-symposium: Pathology of the Prostate
Normal prostatic acinus
Prostatic intra-epithelial neoplasia (PIN)
Localised androgen dependent prostatic adenocarcinoma
Locally advanced androgen dependent prostatic adenocarcinoma
Metastatic androgen dependent prostatic adenocarcinoma
Localised androgen insensitive prostatic adenocarcinoma
Locally advanced androgen insensitive prostatic adenocarcinoma
Metastatic androgen insensitive prostatic adenocarcinoma
Figure 9 The sequence shown in Figure 8 but now with the additional division into androgen-dependent and androgen-independent types.
of screening for prostate cancer would have detected the disease at an earlier stage and the current available screening methods are: measurement of serum PSA; digital rectal examination; and transrectal ultrasonography.11–13 We can now look at the system and define the stages of the disease that would cause an abnormality detectable by these screening methods (Figure 11). This is relatively straightforward but the systems diagram displays it very clearly and enables specific questions about it to be formulated. One such question is whether PIN can cause a raised PSA? The literature on this is equivocal and compounded by the difficulty in imaging and taking biopsies from the prostate. The only way it could be definitively proven would be to have a number of cases where a radical prostatectomy had been performed on patients with a raised PSA in which only PIN, with no invasive cancer, was found by serial sectioning of the entire specimen – clearly not a study that is likely to happen. However, it is theoretically possible that extensive PIN could produce a raised serum PSA so we can indicate this by a dotted extension of the orange line in Figure 11. The influence diagram is also useful for mapping where any therapeutic interventions can occur as we have done in Figure 12. Again this process makes all the information in the system very clear and illustrates a number of points, e.g. that there is a narrow window of opportunity for performing a radical
Normal prostatic acinus
Prostatic intra-epithelial neoplasia (PIN)
rostatectomy or that once a tumour has become androgen p insensitive and metastatic then the therapeutic options are limited to systemic chemotherapy, radiotherapy to bony metastases and bisphosphonate administration.
Phase IV modelling – causal loop diagrams The influence diagram only shows possible influences on elements within the system and the direction of information flow or order in those influences but it does not show whether the effect is to increase or decrease the risk or size of the next element. A new layer of sophistication in systems thinking can be added by including this and these are usually referred to as causal loop diagrams. In these diagrams, if the effect of an influence goes in the same direction as the element it is influencing, e.g. if the first element increases then the second element also increases, then the arrow is marked as ‘s’ for same direction. If the effect is in the other direction then it is marked as ‘o’ for opposite. These directions may alternatively be labelled ‘+’ and ‘−’ if this produces easier comprehension for different audiences. Figure 13 shows the start of a model of the detection and treatment of prostate cancer.
Localised androgen dependent prostatic adenocarcinoma
Locally advanced androgen dependent prostatic adenocarcinoma
Metastatic androgen dependent prostatic adenocarcinoma
Localised androgen insensitive prostatic adenocarcinoma
Locally advanced androgen insensitive prostatic adenocarcinoma
Metastatic androgen insensitive prostatic adenocarcinoma
Figure 10 The diagram from Figure 9 has been taken and a possible pathway of a man who presented with bony metastases from an androgendependent prostate cancer is shown in red.
DIAGNOSTIC HISTOPATHOLOGY 14:3
127
© 2008 Elsevier Ltd. All rights reserved.
Mini-symposium: Pathology of the Prostate
serum prostate specific antigen transrectal ultrasonography digital rectal examination
Normal prostatic acinus
Prostatic intraepithelial neoplasia (PIN)
Localised androgen dependent prostatic adenocarcinoma
Locally advanced androgen dependent prostatic adenocarcinoma
Metastalic androgen dependent prostatic adenocarcinoma
Localised androgen insensitive prostatic adenocarcinoma
Locally advanced androgen insensitive prostatic adenocarcinoma
Metastatic androgen insensitive prostatic adenocarcinoma
Figure 11 The diagram from Figure 9 is overlaid with three different screening modalities showing the stages of prostate cancer that they might be expected to detect. The line for prostate specific antigen (PSA) is shown dotted over prostatic intra-epithelial neoplasia (PIN) because there is controversy as to whether PIN can produce elevated serum PSA.
In this model we have split all the cases of prostate cancer in the whole population into detected and undetected. Obviously it is only the detected cases that will be known to the healthcare system and so can be treated. We have drawn arrows
between the detected and undetected cases and have labelled the arrow going from the undetected cases to the detected with an ‘s’ because if there are more undetected cases in the population then any detection method will produce more detected
androgen ablation/anti-androgen therapy active surveillance brachytherapy radiotherapy to bony metastases
external beam radiotherapy radical prostatectomy
Normal prostatic acinus
Iycopene/selenium rich diet
Prostatic intraepithelial neoplasia (PIN)
Localised androgen dependent prostatic adenocarcinoma
Locally advanced androgen dependent prostatic adenocarcinoma
Metastatic androgen dependent prostatic adenocarcinoma
Localised androgen insensitive prostatic adenocarcinoma
Locally advanced androgen insensitive prostatic adenocarcinoma
Metastatic androgen insensitive prostatic adenocarcinoma
systemic chemotherapy bisphosphonates Figure 12 The diagram from Figure 9 is overlaid with possible therapies for various stages of prostate cancer. It can be seen that the therapeutic window for radical prostatectomy is quite narrow, just cases of localised prostate cancer. For androgen insensitive metastatic prostate cancer the only therapeutic options are systemic chemotherapy and bisphosphonates, shown at the bottom of the diagram. There is reasonable evidence that a diet rich in lycopenes and with sufficient selenium reduces the risk of prostate cancer so this is added to the left-hand side of the diagram.
DIAGNOSTIC HISTOPATHOLOGY 14:3
128
© 2008 Elsevier Ltd. All rights reserved.
Mini-symposium: Pathology of the Prostate
cases. The arrow going from detected to undetected is labelled ‘o’ because the detected cases will be removed from the whole population total of prostate cancer cases. It can be seen that we have now formed a loop between the two elements and if we follow the loop round a few times imagining changes in one of the elements, we can see an emergent behaviour. If the number of undetected cases in the population rises then so will the number of detected cases but because there are now more detected cases, the lower arrow shows us that there will be fewer undetected cases. Another cycle round the loop shows us that since there are now fewer undetected cases there will also be fewer detected cases so the loop is called a balancing loop because any change in one element will be balanced out by changes in other elements in the loop after a few cycles. It may be more familiar as the negative feedback loop that occurs in situations such as biological homeostasis. The loop already tells us important facts about the detection of prostate cancer – because it is a balancing loop then any intervention to detect prostate cancer will always require continuous effort, any increase in detected cases means that it will require more effort to detect the same number of cases in the next cohort, assuming that the detection threshold remains the same. Of course new cases of prostate cancer are always arising, so we add an element to the bottom left of our diagram to indicate this. There are two types of loops that can occur in
S
S
undetected prostate cancer
detected prostate cancer
O new cases of prostate cancer Figure 13 The first elements of a causal loop diagram examining the detection and treatment of prostate cancer. The cases of prostate cancer in a population are divided into those that are undetected (‘silent’) and those that have been detected by any method (e.g. symptomatic presentation, chance detection during transurethral resection for benign prostatic hypertrophy). The arrow going from detected to undetected cases is labelled with an ‘s’ to show that as the number of cases of undetected prostate cancer increase in a population so the number of detected cases will also increase (change in the same direction, hence the ‘s’ label) given an unchanging threshold of detection. The arrow going from detected cases to undetected cases is labelled ‘o’ (for opposite direction) because any increase in detected cases will necessarily cause a decrease in undetected cases. Of course new cases of prostate cancer will arise in the population and these are shown as an element that increases the number of undetected cases in the population.
S
S
locally advanced prostate cancer
S
androgen ablation or anti-androgen therapy
radiotherapy
S chemotherapy
S
S
metastatic prostate cancer
S
S
S
undetected prostate cancer
detected prostate cancer
S
androgen ablation or anti-androgen therapy chemotherapy
S
O new cases of prostate cancer
S S localised prostate cancer
S
S
radical prostatectomy
radiotherapy active surveillance
Figure 14 The different stages of prostate cancer (shown in blue) have been added to the side of the detected cases element from Figure 13. This is in effect the addition of the elements from Figure 12 added to Figure 13 but as elements in a causal loop diagram rather than as a flow diagram. However, there is little interaction between the elements as these additions just map the currently available treatments (shown in purple) for different stages of prostate cancer. All the arrows are labelled as ‘s’ because any increase in the number of detected cases of prostate cancer will cause any increase in the numbers of cases in each stage and the different treatments for that stage. The relative split of these numbers is not shown and is not initially necessary for the model. There are two arrows between the different stages showing that without treatment localised and locally advanced prostate cancer can progress to metastatic prostate cancer (and so the arrows are labelled ‘s’).
DIAGNOSTIC HISTOPATHOLOGY 14:3
129
© 2008 Elsevier Ltd. All rights reserved.
Mini-symposium: Pathology of the Prostate
S
S
locally advanced prostate cancer
S
androgen ablation or anti-androgen therapy radiotherapy
S chemotherapy S S undetected prostate cancer S
new cases of prostate cancer
detected prostate cancer
S
S S
metastatic prostate cancer S
O
O
O
S
death from S prostate cancer
S chemotherapy O
S S localised prostate cancer
androgen ablation or anti-androgen therapy
S
S S
radical prostatectomy radiotherapy active surveillance
Figure 15 This is Figure 14 with the addition of outcomes (shown in green). The only outcome in this diagram that is shown is death from prostate cancer. Obviously in a full model we would have to include death from other causes, death from treatment of prostate cancer and quality of life, but in this initial model we are examining the effects of increased detection of prostate cancer on deaths from prostate cancer so we omit the other elements for clarity. We have put arrows labelled ‘s’ on all the states where a treatment is not curative, e.g. androgen ablation therapy for locally advanced prostate cancer (although that therapy does prolong life). The long loop from undetected prostate cancer to death from prostate cancer is an important one that might be difficult to detect unless an autopsy was performed. We add three arrows labelled ‘o’ from the three treatments for localised prostate cancer to metastatic prostate cancer because we know that in many cases those treatments are curative.
causal loop diagrams – balancing and reinforcing. A reinforcing loop will be more familiar as a positive feedback loop where a small change in one element can precipitate a sequence of events with large changes in many other elements. Reinforcing loops will not appear very often, if at all, in systems modelling human cancer but would appear with more frequency in models of infectious disease in humans when they would be the mechanism that would generate an epidemic. We can now add the different stages of detected cases of prostate cancer, their possible treatments and the possible movements between stages (Figure 14). These are all just straightforward positive flows of cases through a branching network and there are no loops within it.
DIAGNOSTIC HISTOPATHOLOGY 14:3
It is only when we add outcomes from the treatment that the diagram becomes more interesting (Figure 15). We know that there is currently no curative treatment for metastatic prostate cancer (androgen ablation therapy and chemotherapy prolong survival but are not curative), so any patient with metastatic prostate cancer is quite likely to die from that disease. We know that androgen ablation/anti-androgen therapy is effective for androgen-sensitive tumours but that tumours eventually become androgen insensitive so patients treated by this alone are likely to die of the disease. Undetected cases of prostate cancer in the population may cause death from prostate cancer that is only detected at autopsy so we can put a long arrow from that side of the diagram to death from prostate cancer. All these 130
© 2008 Elsevier Ltd. All rights reserved.
Mini-symposium: Pathology of the Prostate
S androgen ablation or anti-androgen therapy serum PSA screening S
locally advanced prostate cancer
S
radiotherapy
S chemotherapy S S undetected prostate cancer S
new cases of prostate cancer
S
detected prostate cancer
S
S S
metastatic prostate cancer
S
O
O
S
O
O
S
death from S prostate cancer
chemotherapy
radical prostatectomy
S S localised prostate cancer
androgen ablation or anti-androgen therapy
S
S S
radiotherapy active surveillance
Figure 16 We have taken Figure 15 and added the element of prostate screening by serum prostate specific antigen (PSA), which is linked to identified prostate cancer by an arrow labelled ‘s’ because we know it increases the detection rate of prostate cancer. We have then highlighted in red the pathways that would reduce deaths from prostate cancer that could be increased by prostate cancer screening – i.e. detection of localised prostate cancer that is then treated, thereby reducing the number of cases of metastatic prostate cancer. This is just to illustrate what would need to happen for PSA screening to reduce deaths from prostate cancer. The diagram as a whole shows that the number of cases in all the other stages would also be increased – a relative increase in detection of cases of localised prostate cancer would be needed to make PSA screening effective. The diagram also highlights the sorts of problematic cases that could be thrown up by screening – e.g. an asymptomatic man who is discovered to have metastatic prostate cancer.
arrows are labelled ‘s’ because they will increase the number of deaths from prostate cancer. On the benefit side of the equation we know that radical prostatectomy performed on localised prostate cancer is very often curative and so will reduce the number of cases of metastatic prostate cancer and subsequent deaths from prostate cancer. Radiotherapy on localised prostate cancer has a similar effect. Active surveillance of localised prostate cancer should also reduce the cases of metastatic prostate cancer although, to be totally accurate in the diagram, there will be some cases that appear to be progressing, which will require intervention by radical prostatectomy or radiotherapy to produce this effect. Now that we have created a simple model of prostate cancer and its treatment we can look at the effect of an intervention such as prostate cancer screening by serum PSA (Figure 16).
DIAGNOSTIC HISTOPATHOLOGY 14:3
The only pathway that will reduce the number of deaths from prostate cancer in the population is shown in red. The screening will cause more cases of prostate cancer to be detected in the population and, of these, a proportion will be localised that can be treated to prevent the development of metastatic prostate cancer with a subsequent reduction in deaths from prostate cancer. However, we can also see that the process of screening will also increase (at least on an initial screen) the number of detected cases with locally advanced and metastatic prostate cancer and it could be a challenge to decide how to manage these cases. If a man presented to his general practitioner with severe backache and was found to have bony metastases from prostate cancer, then the decision to treat him with androgen ablation/anti-androgen therapy would be clear-cut because he had symptoms from the disease that were reducing his quality of life. If a man was 131
© 2008 Elsevier Ltd. All rights reserved.
Mini-symposium: Pathology of the Prostate
found to have bony metastases from prostate cancer at screening but did not have any symptoms, then the decision to treat would be more difficult: we know that androgen ablation/anti-androgen therapy has only a limited period of effectiveness before the tumour becomes androgen insensitive so it might be prudent to wait until the man has developed symptoms before starting this treatment. However, the screening process has produced a man who thought he was well but has now found that he has cancer, which has spread through his body but the doctors who detected it don’t want to treat it at the moment – a difficult situation to explain to many patients!14 By looking at this simple model and following the different pathways that it presents, many such scenarios can be developed that might not have been apparent before the model was constructed. All these scenarios can be considered in a qualitative and semi-quantitative way before any decisions about whether or not to introduce prostate screening are made. The model can also be used as a basis for discussion between all parties involved in such decisions because it presents the information in such a clear and holistic manner. Any particular sector of the healthcare system can look at the model and assess the effects that any intervention will have on their work. Histopathologists could look at screening by PSA as an intervention and could predict that it would increase the number of prostate core biopsies and radical prostatectomy specimens that would be processed in their laboratories. They could also predict that there was unlikely to be any reduction in their workload from changes in other parts of the system because biopsies are rarely taken from metastatic prostate cancer. All of these analyses and predictions can be made from a simple model that would only take a few hours at most to generate from knowledge that healthcare staff already have. It might not be a perfect model, and different healthcare groups might derive different models for the same situation, but it will produce informed and holistic discussion about any interventions in the system that is likely to improve the quality of any subsequent decision-making.
using values of elements in their mid-range but perform badly at the extremes of these ranges (which still need to be tested, e.g. what is the value of prostate cancer screening if its incidence was 10 times less than its current level?). There are some methods that can speed up the development of quantitative models and produce reasonable estimates of values where there are several unknowns. There are some proprietary software programmes that enable causal loop diagrams to be encoded using a simple interface, and quantitative values to be entered, which will at least produce reasonable predictions of trends.15,16 These are valuable when a system becomes too complex to make predictions from simple visual inspection.
Conclusions A systems approach to prostate cancer can produce simple models that give insight into many aspects of its pathogenesis, diagnosis and management. The very act of creation of these models makes the creators focus on the available information in an analytical and holistic way, which is likely to enhance their understanding of the situation. Any models that they produce will be useful for discussing the situation with other stakeholders and could save a huge amount of time and, possibly, misdirected resources. The optimal level of payoff for model development in most medical environments is probably the causal loop diagram, which takes relatively little time to create but can be used to generate many scenarios and predictions. Although fully quantitative models are intrinsically attractive to many scientifically-minded healthcare staff, the resources that are required to develop and validate them and the intrinsic problems of making quantitative predictions from incomplete data make them less immediately useful. Nevertheless, such models remain a great challenge for the future. ◆
References 1 Sherwood D. Seeing the forest for the trees – a manager’s guide to applying systems thinking. London: Nicholas Brealey, 2002. 2 Cavana RY, Mares ED. Integrating critical thinking and systems thinking: from premises to causal loops. Syst Dyn Rev 2004; 20: 223–35. 3 Maani KE, Maharaj V. Links between systems thinking and complex decision making. Syst Dyn Rev 2004; 20: 21–48. 4 Wolstenholme E. Using generic system archetypes to support thinking and modelling. Syst Dyn Rev 2004; 20: 341–56. 5 Brawer MK. Prostatic intraepithelial neoplasia: an overview. Rev Urol 2005; 7: S11–18. 6 Iczkowski KA. Current prostate biopsy interpretation – criteria for cancer, atypical small acinar proliferation, high-grade prostatic intraepithelial neoplasia, and use of immunostains. Arch Pathol Lab Med 2006; 130: 835–43. 7 Ayala AG, Ro JY. Prostatic intraepithelial neoplasia – recent advances. Arch Pathol Lab Med 2007; 131: 1257–66. 8 Bostwick DG, Meiers I. Atypical small acinar proliferation in the prostate. Arch Pathol Lab Med 2006; 130: 952–7. 9 Richter E, Srivastava S, Dobi A. Androgen receptor and prostate cancer. Prostate Cancer Prostatic Dis 2007; 10: 114–8. 10 Grossmann ME, Huang H, Tindall DJ. Androgen receptor signaling in androgen-refractory prostate cancer. J Natl Cancer Inst 2001; 93: 1687–97.
Phase V modelling – quantitative models A quantitative model is often viewed as the Holy Grail of systems modelling. If the model in Figure 16 could be turned into a model where numbers could be entered into the various elements, then the effects of interventions could be quantified, e.g. the number of localised prostate cancers that would be detected from one round of PSA prostate cancer screening in men over 50 years in the UK. It is possible to produce such quantitative models but the required resources are orders of magnitude more than that needed to create a sophisticated causal loop diagram. The data for all the elements must to be obtained from reliable sources, usually from meta-analyses of randomised controlled clinical trials. The dynamics of each interaction in the system needs to be clearly and accurately defined and then encoded in an appropriate mathematical formula. A project that would produce a gold standard quantitative model of Figure 16 would require many months of work. There are other problems with quantitative models – any small errors in one part of the system are likely to be magnified (especially by reinforcing loops) in other parts of the system and may produce completely misleading results. Conventional quantitative models often given reasonable results
DIAGNOSTIC HISTOPATHOLOGY 14:3
132
© 2008 Elsevier Ltd. All rights reserved.
Mini-symposium: Pathology of the Prostate
11 Thompson IM, Ankerst DP. Prostate-specific antigen in the early detection of prostate cancer. CMAJ 2007; 176: 1853–8. 12 Fitzsimons NJ, Sun L, Moul JW. Medical technologies for the diagnosis of prostate cancer. Expert Rev Med Devices 2007; 4: 227–39. 13 Ilic D, O’Connor D, Green S, Wilt T. Screening for prostate cancer: a Cochrane systematic review. Cancer Causes Control 2007; 18: 279–85. 14 Rimer BK, Briss PA, Zeller PK, Chan ECY, Woolf SH. Informed decision making: what is its role in cancer screening? Cancer 2004; 101(suppl 5): 1214–28. 15 iThink, Stella. isee systems inc., Lebanon, New Hamphire, USA. Available from: http://www.iseesystems.com/ [accessed 26.02.08] [iThink and Stella are the same software product but the literature that accompanies them is slanted to business applications (in iThink) and education/research (in Stella)]. 16 Vensim. Ventana Systems, Inc., Harvard, Massachusetts, USA. Available from: http://www.vensim.com/ [accessed 26.02.08].
DIAGNOSTIC HISTOPATHOLOGY 14:3
Practice points • A systems modelling approach provides useful insights into the pathogenesis, diagnosis and management of prostate cancer • State and influence diagrams provide a clear mapping of current situations that can be useful for explaining diagnostic and treatment protocols • Causal loop diagrams are very useful for making qualitative and semi-quantitative predictions about any changes in a system of diagnosis and/or management of prostate cancer • Quantitative hard systems models take a huge amount of resource to develop and validate, and should only be used when softer methodologies have been fully explored
133
© 2008 Elsevier Ltd. All rights reserved.
Prostate cancer: a systems approach overview
more difficult and has less resolution than direct inspection. This leads to fewer biopsies and less participation of histopathologists in the whole spectrum of a disease. This restriction of biopsy material to a few points along the disease spectrum can lead to a distortion of histopathologists’ perception of the disease and some inflexibility when faced with changes in management of that disease, e.g. introduction of a cancer screening programme. A systems approach, often called ‘systems thinking’, is a methodology that sets out to create an overall model of a particular situation, which can be used to make predictions about that situation if specific elements within it are changed. The systems approach is often divided into two categories – soft and hard systems. A soft systems approach identifies all the elements within a system and the qualitative relationships between these. A hard systems approach goes further, taking quantitative data and producing a model that will produce quantitative predictions about outcomes from changes in input data. A hard systems approach will be more familiar to scientists since much translational molecular pathology research contains an element of hard systems modelling, e.g. development of a novel molecular marker that will predict systemic metastases in node negative breast cancer. The soft systems approach has had less application in the scientific and biomedical arenas and is often derided in those areas because it lacks a quantitative basis. However, soft systems methodologies have had widespread application in management ‘science’ where they have been shown to be very successful at addressing complex problems.1–4 Since diagnostic histopathology is often a blend of scientific knowledge integrated within a management context, it is likely that a soft systems approach could yield valuable insights into the effects of change in the scientific knowledge or the management of diseases. This review applies some systems approaches to prostate cancer and shows how these can focus our thoughts on the scientific facts about the disease as well as enabling us to predict how changes in management of the disease will affect histopathology services. The review starts with overly simplistic models but then builds up to more complex models that should have some clinical utility.
Simon S Cross Freddie C Hamdy John R Goepel Robert F Harrison
Abstract Systems thinking is a set of methodologies that facilitate analysis and predictions for a complex system in a holistic way. Prostate cancer has precursor stages that are difficult to detect and a number of different therapy options for different stages, so it is a complex disease to manage within healthcare systems. In this review we show how systems thinking, especially causal loop diagrams, can be a very valuable tool for gaining greater insight into the pathogenesis and diagnosis of prostate cancer and to predict the consequences in major changes in the pattern of healthcare for this disease, e.g. introduction of national screening programmes using serum prostate specific antigen. The systems thinking approach can be used to predict changes in histopathology workflow when other parts of the system are changed.
Keywords causal loop diagrams; influence diagrams; prostate cancer; prostate cancer screening; prostatic intra-epithelial neoplasia; prostate specific antigen; state diagrams; systems thinking
Introduction Histopathology has an intermittent relationship with the management of many diseases because the opportunities for biopsy and histopathological examination are often limited to a few points along the disease progression. Some organs, the colorectum being an exemplar, are relatively accessible for visual inspection and biopsy so a disease such as colorectal cancer is well-represented in histopathology biopsies from early precursor lesions through to advanced disease. Other organs, including the prostate, are not available for visual inspection and the imaging and biopsy of these organs is
Phase I modelling – a simple state diagram The simplest level of a systems approach is to identify the elements within a system without ordering them or defining any relationships between them. This may appear completely selfevident, and when one has previous knowledge of the subject it may appear unnecessary, but it is important to do this with as much inclusion as possible so that no potentially important elements are left out of the system. It becomes extremely important to do this when trying to model a new system about which there is very little initial information. If we think about prostate cancer at its most basic level we know that men are born with (we assume) normal prostates but that a sizeable proportion of men develop prostate cancer so at its very simplest level we can identify two states – normal and prostate cancer (Figure 1).
Simon S Cross MD FRCPath is at the Academic Unit of Pathology, School of Medicine & Biomedical Sciences, University of Sheffield, UK. Freddie C Hamdy FRCS is at the Academic Unit of Urology, School of Medicine & Biomedical Sciences, University of Sheffield, UK. John R Goepel FRCPath is at the Department of Histopathology, Royal Hallamshire Hospital, Sheffield Teaching Hospitals NHS Foundation Trust, Sheffield UK.
Normal prostate
Robert F Harrison Department of Automatic Control & Systems Engineering, University of Sheffield, UK.
DIAGNOSTIC HISTOPATHOLOGY 14:3
Prostate cancer
Figure 1 A simple two state representation of normality and prostate cancer.
122
© 2008 Elsevier Ltd. All rights reserved.
Mini-symposium: Pathology of the Prostate
Of course we know that there are many more entities in the prostate cancer spectrum and so we can define those to give more elements in our state diagram (Figure 2). This state diagram has no order to the elements, they are simply put onto the page in the order that we have thought of them, which would seem to be unhelpful but it is worth reflecting that histopathology when performed in an uncritical manner would be somewhat analogous to such a diagram. In histopathology we are taught to identify reliably discrete morphological patterns and to assign them agreed labels, which are included in our reports and should be understood by clinicians reading those reports. Diagnostic histopathology can run at this most basic level without much detriment to the overall management of patients as long as the labels can be applied to the morphological patterns with a one-to-one correspondence and a high level of reliability. However, it is much better if the elements are more ordered with clear definitions of their relationships because this gives a much greater understanding of the interaction between diagnostic histopathology and overall patient management, especially in difficult areas, e.g. atypical small acinar proliferation and early prostate cancer in needle core biopsies. Looking at the simple state diagram in Figure 2 there are still some important points that arise. Among the entities included are neuroendocrine prostate cancer and large duct prostate cancer (shown in red in Figure 3). When we consider those in the context of the other elements we realise that, although they are variants of prostate cancer, they are not going to fit in with the overall progression that we are going to construct. This will be based on the most common adenocarcinoma of the prostate that arises from the acini in the prostate with no special pattern of differentiation. These other entities would need to be included in a complete systems model of prostate cancer but are likely to add unnecessary complexity to our initial model so we will omit them. It also means that we need to tighten up on our terminology for the existing elements so we end the first phase of modelling with the state diagram in Figure 4.
Phase II modelling – an influence/progression diagram Now that we have defined elements that we wish to include in the model, we can look at the relationships between these elements and indicate them on the diagram by arrows. When we look at the later stages of prostate cancer the elements are relatively easy to order. Localised prostate cancer can progress to locally advanced prostate cancer, which could, in turn, progress to metastatic prostate cancer as shown in Figure 5. Many diagrams like this have been reproduced in the pathology literature, Walter Bodmer’s progression from colorectal adenoma through to invasive colorectal carcinoma being the prototypical example. It is important to get all the details of such sequences correct as such diagrams spread rapidly through the literature, presumably owing to the much easier assimilation of information from diagrams rather than blocks of dense text. Looking at our new diagram (Figure 5) we need to look at all the elements and ensure that all appropriate arrows have been included to indicate all possible relationships. We ask ourselves whether prostate cancer can metastasise from localised, as well as locally advanced, prostate cancer. That can indeed occur so we need to draw an arrow from localised to metastatic prostate cancer, which does not pass through locally advanced prostate cancer as an obligate transitional stage (Figure 6). We can also look at whether the arrows should be one way or two way, i.e. can locally advanced prostate cancer regress to localised or metastatic prostate cancer regress to localised prostate cancer? The answers to these questions are obviously no, at least without some very effective therapy, so the arrows should be left one way. Including the precursor lesions to invasive prostate cancer in the influence diagram is more difficult and this is where the value of these diagrams becomes evident. Prostatic intra-epithelial neoplasia (PIN) is probably the easier of the two to deal with. What we need to know is whether PIN is a precursor from which invasive prostate cancer develops and if so is it an obligate precursor,
Neuroendocrine prostate cancer
Low grade prostatic intraepithelial neoplasia (PIN)
Normal prostate
Atypical small acinar proliferation
Locally advanced prostate cancer
Localised prostate cancer
Metastatic prostate cancer
Large duct prostate cancer
High grade prostatic intraepithelial neoplasia (PIN) Figure 2 A state diagram showing all the elements that might be considered for a model of prostate cancer pathogenesis and progression. The elements are not ordered in any way.
DIAGNOSTIC HISTOPATHOLOGY 14:3
123
© 2008 Elsevier Ltd. All rights reserved.
Mini-symposium: Pathology of the Prostate
Neuroendocrine prostate cancer
Low grade prostatic intraepithelial neoplasia (PIN)
Normal prostate
Large duct prostate cancer
Metastatic prostate cancer
Atypical small acinar proliferation
Locally advanced prostate cancer
Localised prostate cancer
High grade prostatic intraepithelial neoplasia (PIN) Figure 3 The state diagram from Figure 2 has been taken and two elements, neuroendocrine prostate cancer and large duct cancer, have been highlighted in red because they do not fit into a general system of the most common non-specialised prostate cancer.
i.e. do all cases of ordinary prostatic adenocarcinoma arise from foci of PIN. If PIN is not a precursor but only a marker of risk of development of prostate cancer then it will not be included in the direct disease progression sequence. If it is a precursor but not an obligate one, then it will be in the sequence but there will be an alternative parallel route. These options are illustrated in Figure 7. The data to decide between these different options is quite difficult to acquire for the prostate because of the previously discussed
problems with imaging and taking targeted biopsies from the organ. In the colorectum the picture is much clearer because abnormalities can be seen at an almost unicryptal level as aberrant crypt foci using magnifying chromoendoscopy and can even be subtyped into hyperplastic and adenomatous aberrant crypt foci. Thus, the current theory in colorectal cancer is that aberrant crypt foci are obligate precursors of colorectal cancer. In the prostate such small precursor lesions cannot be identified in vivo and the main source
Low grade prostatic intraepithelial neoplasia (PIN)
Normal prostatic acinus
Atypical small acinar proliferation
Localised prostatic adenocarcinoma
Locally advanced prostatic adenocarcinoma
Metastatic prostatic adenocarcinoma
High grade prostatic intraepithelial neoplasia (PIN) Figure 4 The state diagram from Figure 3 has been taken, the two highlighted elements have been omitted and the terminology in the other elements has been altered to be more specific, e.g. prostatic adenocarcinoma rather than prostate cancer.
DIAGNOSTIC HISTOPATHOLOGY 14:3
124
© 2008 Elsevier Ltd. All rights reserved.
Mini-symposium: Pathology of the Prostate
Localised prostatic adenocarcinoma
Locally advanced prostatic adenocarcinoma
Metastatic prostatic adenocarcinoma
Figure 5 The elements in the later stages of prostate cancer have been ordered into a sequence.
suboptimal histological processing.8 It is, therefore, used as a histological term that suggests rebiopsy in order to obtain sufficient histological material to make a definitive diagnosis of invasive prostate cancer and is thus not a biological precursor of prostate cancer. One piece of evidence that provides scant evidence that it could be a precursor, or at least a true biological marker of risk, is that if prostates are rebiopsied following a diagnosis of atypical small acinar proliferation in unilateral biopsies, then in 39% of cases the invasive cancer was found exclusively in a site distant from the initial biopsies.8 This is confusing since if atypical small acinar proliferation was simply a term relating to invasive cancer that could not be diagnosed in the particular biopsy under consideration, then one would expect cancers diagnosed immediately after the first sample to be at the same site as the initial biopsies. However, on current published evidence and definitions, atypical small acinar proliferation cannot be included as a bona fide biological precursor of invasive prostate cancer. Figure 8 now looks like a reasonable summary of the progression of prostate cancer but we need to include one more factor in the later stages of the disease – androgen dependency.9 This is a factor that we rarely encounter in histopathology because we do not test prostate cancers for androgen receptors in the same way that we do breast cancers for oestrogen and progesterone receptors. It is, however, a vital factor in determining treatment of non-localised prostate cancer and so needs to be included in our system. We have included this in Figure 9 on the basis that, when it arises, prostate cancer will be androgen sensitive but at any point after that it may become androgen insensitive whether by spontaneous mutations within the tumour in an untreated patient or by selection of androgen-insensitive clones of the tumour in patients receiving androgen ablation therapy. We also know that androgen-insensitive tumours do not become androgen dependent again by further accumulated mutations, so the arrows are uni-directional.10
of information is from painstaking examination of radical prostatectomy specimens where the main identified invasive prostatic adenocarcinoma is likely to have outgrown and destroyed any precursor lesions. The current published literature strongly suggests that PIN is a direct precursor of prostate cancer.5–7 It is found in the same areas of the prostate as invasive cancer with the same relative frequency as prostate cancer in different groups, i.e. increasing incidence with increasing age and increased incidence in groups with known increased incidence of prostate cancer. Many of the molecular abnormalities of prostate cancer are also found in PIN but not in the morphologically normal background acini in prostates that contain cancer. No conclusive evidence appears to exist to suggest that all prostate cancer arises from PIN; this may be due to the problems with detection but means that, at present, an alternative parallel pathway has to be included. In Figure 4 we split PIN into low and high grade because these are morphological categories used in diagnostic histopathology. We know that we only include high grade PIN in our histopathology reports because the intra- and interobserver variability for this diagnosis is within reasonable limits whereas the same parameters for low grade PIN are not sufficiently reliable to act as a basis for any clinical decisions. However, if we are looking at PIN in the context of the spectrum of prostate cancer, it is more logical to include it as a single category rather than two rather arbitrary morphological categories. The final feature that we need to consider is whether PIN can regress, i.e. whether the arrow between normal prostatic acini and PIN should be one or two way? There is considerable evidence in the literature that PIN can regress under treatment conditions, such as androgen ablation therapy, and some evidence that lesser changes in the prostatic milieu, such as selenium or vitamin E as dietary supplements, may also induce regression – so a two-way arrow or separate arrows in opposite directions seems reasonable (Figure 8).5–7 The place of atypical small acinar proliferation as a precursor of prostate cancer is less secure. The published definitions of this term imply that it is a histological appearance that is suspicious of invasive prostatic adenocarcinoma but that the definitive diagnosis cannot be made because of some compromising features of the biopsy such as the small size of the focus of abnormality, some abnormal features of morphology, e.g. atrophy, or artefacts from
Localised prostatic adenocarcinoma
Phase III modelling – using the influence diagram Now that we have constructed a fairly comprehensive influence diagram for the whole spectrum of prostate cancer, we can use this
Locally advanced prostatic adenocarcinoma
Metastatic prostatic adenocarcinoma
Figure 6 An additional loop has been added to the diagram from Figure 5 because it is recognised that localised prostatic adenocarcinoma can metastasise without necessarily passing through a locally advanced stage.
DIAGNOSTIC HISTOPATHOLOGY 14:3
125
© 2008 Elsevier Ltd. All rights reserved.
Mini-symposium: Pathology of the Prostate
a
PIN is an obligate precursor of prostatic adenocarcinoma
Normal prostatic acinus
b
Localised prostatic adenocarcinoma
PIN is a non-obligate precursor of prostatic adenocarcinoma
Normal prostatic acinus
c
Prostatic intraepithelial neoplasia (PIN)
Prostatic intraepithelial neoplasia (PIN)
Localised prostatic adenocarcinoma
PIN is not a precursor of prostatic adenocarcinoma
Localised prostatic adenocarcinoma
Normal prostatic acinus
Prostatic intraepithelial neoplasia (PIN) Figure 7 Three different possible roles of prostatic intra-epithelial neoplasia (PIN) in the pathogenesis of prostate cancer are shown. a PIN is an obligate precursor of b PIN is a non-obligate precursor of prostatic adenocarcinoma, i.e. some but not all cases of prostatic adenocarcinoma rise from PIN. c PIN is not a precursor of prostatic adenocarcinoma, i.e. PIN is a marker of increased risk of development of prostatic adenocarcinoma but it does not arise from it. Note how the diagrams summarise these possibilities much more succinctly than text.
in our thinking about the disease. It is easy and instructive to plot an individual patient’s pathway through the system to the point at which they present to the healthcare system. Figure 10 shows the pathway of a man who presented to his general practitioner with back pain and was found to have sclerotic bone metastases in his lumbar spine and a raised serum prostate specific antigen (PSA). What we can see immediately (and which would have to be determined from some clinical information, such as response to androgen ablation therapy) is that, although metastatic, his
Normal prostatic acinus
Prostatic intraepithelial neoplasia (PIN)
tumour is androgen dependent so he should have a response to androgen ablation therapy. Since he is clearly presenting at a late stage of disease and will have been through many earlier stages, then it is instructive to see whether any healthcare interventions made earlier would have prevented such a late stage presentation. If he had no symptoms of prostatism (and he need not have since most prostate cancers are relatively small and peripheral) then he would have had no cause to present to the healthcare system until the onset of his back pain. This means that only some form
Localised prostatic adenocarcinoma
Locally advanced prostatic adenocarcinoma
Metastatic prostatic adenocarcinoma
Figure 8 The complete sequence of prostate cancer pathogenesis and progression from normal prostate through to metastatic disease.
DIAGNOSTIC HISTOPATHOLOGY 14:3
126
© 2008 Elsevier Ltd. All rights reserved.
Mini-symposium: Pathology of the Prostate
Normal prostatic acinus
Prostatic intra-epithelial neoplasia (PIN)
Localised androgen dependent prostatic adenocarcinoma
Locally advanced androgen dependent prostatic adenocarcinoma
Metastatic androgen dependent prostatic adenocarcinoma
Localised androgen insensitive prostatic adenocarcinoma
Locally advanced androgen insensitive prostatic adenocarcinoma
Metastatic androgen insensitive prostatic adenocarcinoma
Figure 9 The sequence shown in Figure 8 but now with the additional division into androgen-dependent and androgen-independent types.
of screening for prostate cancer would have detected the disease at an earlier stage and the current available screening methods are: measurement of serum PSA; digital rectal examination; and transrectal ultrasonography.11–13 We can now look at the system and define the stages of the disease that would cause an abnormality detectable by these screening methods (Figure 11). This is relatively straightforward but the systems diagram displays it very clearly and enables specific questions about it to be formulated. One such question is whether PIN can cause a raised PSA? The literature on this is equivocal and compounded by the difficulty in imaging and taking biopsies from the prostate. The only way it could be definitively proven would be to have a number of cases where a radical prostatectomy had been performed on patients with a raised PSA in which only PIN, with no invasive cancer, was found by serial sectioning of the entire specimen – clearly not a study that is likely to happen. However, it is theoretically possible that extensive PIN could produce a raised serum PSA so we can indicate this by a dotted extension of the orange line in Figure 11. The influence diagram is also useful for mapping where any therapeutic interventions can occur as we have done in Figure 12. Again this process makes all the information in the system very clear and illustrates a number of points, e.g. that there is a narrow window of opportunity for performing a radical
Normal prostatic acinus
Prostatic intra-epithelial neoplasia (PIN)
rostatectomy or that once a tumour has become androgen p insensitive and metastatic then the therapeutic options are limited to systemic chemotherapy, radiotherapy to bony metastases and bisphosphonate administration.
Phase IV modelling – causal loop diagrams The influence diagram only shows possible influences on elements within the system and the direction of information flow or order in those influences but it does not show whether the effect is to increase or decrease the risk or size of the next element. A new layer of sophistication in systems thinking can be added by including this and these are usually referred to as causal loop diagrams. In these diagrams, if the effect of an influence goes in the same direction as the element it is influencing, e.g. if the first element increases then the second element also increases, then the arrow is marked as ‘s’ for same direction. If the effect is in the other direction then it is marked as ‘o’ for opposite. These directions may alternatively be labelled ‘+’ and ‘−’ if this produces easier comprehension for different audiences. Figure 13 shows the start of a model of the detection and treatment of prostate cancer.
Localised androgen dependent prostatic adenocarcinoma
Locally advanced androgen dependent prostatic adenocarcinoma
Metastatic androgen dependent prostatic adenocarcinoma
Localised androgen insensitive prostatic adenocarcinoma
Locally advanced androgen insensitive prostatic adenocarcinoma
Metastatic androgen insensitive prostatic adenocarcinoma
Figure 10 The diagram from Figure 9 has been taken and a possible pathway of a man who presented with bony metastases from an androgendependent prostate cancer is shown in red.
DIAGNOSTIC HISTOPATHOLOGY 14:3
127
© 2008 Elsevier Ltd. All rights reserved.
Mini-symposium: Pathology of the Prostate
serum prostate specific antigen transrectal ultrasonography digital rectal examination
Normal prostatic acinus
Prostatic intraepithelial neoplasia (PIN)
Localised androgen dependent prostatic adenocarcinoma
Locally advanced androgen dependent prostatic adenocarcinoma
Metastalic androgen dependent prostatic adenocarcinoma
Localised androgen insensitive prostatic adenocarcinoma
Locally advanced androgen insensitive prostatic adenocarcinoma
Metastatic androgen insensitive prostatic adenocarcinoma
Figure 11 The diagram from Figure 9 is overlaid with three different screening modalities showing the stages of prostate cancer that they might be expected to detect. The line for prostate specific antigen (PSA) is shown dotted over prostatic intra-epithelial neoplasia (PIN) because there is controversy as to whether PIN can produce elevated serum PSA.
In this model we have split all the cases of prostate cancer in the whole population into detected and undetected. Obviously it is only the detected cases that will be known to the healthcare system and so can be treated. We have drawn arrows
between the detected and undetected cases and have labelled the arrow going from the undetected cases to the detected with an ‘s’ because if there are more undetected cases in the population then any detection method will produce more detected
androgen ablation/anti-androgen therapy active surveillance brachytherapy radiotherapy to bony metastases
external beam radiotherapy radical prostatectomy
Normal prostatic acinus
Iycopene/selenium rich diet
Prostatic intraepithelial neoplasia (PIN)
Localised androgen dependent prostatic adenocarcinoma
Locally advanced androgen dependent prostatic adenocarcinoma
Metastatic androgen dependent prostatic adenocarcinoma
Localised androgen insensitive prostatic adenocarcinoma
Locally advanced androgen insensitive prostatic adenocarcinoma
Metastatic androgen insensitive prostatic adenocarcinoma
systemic chemotherapy bisphosphonates Figure 12 The diagram from Figure 9 is overlaid with possible therapies for various stages of prostate cancer. It can be seen that the therapeutic window for radical prostatectomy is quite narrow, just cases of localised prostate cancer. For androgen insensitive metastatic prostate cancer the only therapeutic options are systemic chemotherapy and bisphosphonates, shown at the bottom of the diagram. There is reasonable evidence that a diet rich in lycopenes and with sufficient selenium reduces the risk of prostate cancer so this is added to the left-hand side of the diagram.
DIAGNOSTIC HISTOPATHOLOGY 14:3
128
© 2008 Elsevier Ltd. All rights reserved.
Mini-symposium: Pathology of the Prostate
cases. The arrow going from detected to undetected is labelled ‘o’ because the detected cases will be removed from the whole population total of prostate cancer cases. It can be seen that we have now formed a loop between the two elements and if we follow the loop round a few times imagining changes in one of the elements, we can see an emergent behaviour. If the number of undetected cases in the population rises then so will the number of detected cases but because there are now more detected cases, the lower arrow shows us that there will be fewer undetected cases. Another cycle round the loop shows us that since there are now fewer undetected cases there will also be fewer detected cases so the loop is called a balancing loop because any change in one element will be balanced out by changes in other elements in the loop after a few cycles. It may be more familiar as the negative feedback loop that occurs in situations such as biological homeostasis. The loop already tells us important facts about the detection of prostate cancer – because it is a balancing loop then any intervention to detect prostate cancer will always require continuous effort, any increase in detected cases means that it will require more effort to detect the same number of cases in the next cohort, assuming that the detection threshold remains the same. Of course new cases of prostate cancer are always arising, so we add an element to the bottom left of our diagram to indicate this. There are two types of loops that can occur in
S
S
undetected prostate cancer
detected prostate cancer
O new cases of prostate cancer Figure 13 The first elements of a causal loop diagram examining the detection and treatment of prostate cancer. The cases of prostate cancer in a population are divided into those that are undetected (‘silent’) and those that have been detected by any method (e.g. symptomatic presentation, chance detection during transurethral resection for benign prostatic hypertrophy). The arrow going from detected to undetected cases is labelled with an ‘s’ to show that as the number of cases of undetected prostate cancer increase in a population so the number of detected cases will also increase (change in the same direction, hence the ‘s’ label) given an unchanging threshold of detection. The arrow going from detected cases to undetected cases is labelled ‘o’ (for opposite direction) because any increase in detected cases will necessarily cause a decrease in undetected cases. Of course new cases of prostate cancer will arise in the population and these are shown as an element that increases the number of undetected cases in the population.
S
S
locally advanced prostate cancer
S
androgen ablation or anti-androgen therapy
radiotherapy
S chemotherapy
S
S
metastatic prostate cancer
S
S
S
undetected prostate cancer
detected prostate cancer
S
androgen ablation or anti-androgen therapy chemotherapy
S
O new cases of prostate cancer
S S localised prostate cancer
S
S
radical prostatectomy
radiotherapy active surveillance
Figure 14 The different stages of prostate cancer (shown in blue) have been added to the side of the detected cases element from Figure 13. This is in effect the addition of the elements from Figure 12 added to Figure 13 but as elements in a causal loop diagram rather than as a flow diagram. However, there is little interaction between the elements as these additions just map the currently available treatments (shown in purple) for different stages of prostate cancer. All the arrows are labelled as ‘s’ because any increase in the number of detected cases of prostate cancer will cause any increase in the numbers of cases in each stage and the different treatments for that stage. The relative split of these numbers is not shown and is not initially necessary for the model. There are two arrows between the different stages showing that without treatment localised and locally advanced prostate cancer can progress to metastatic prostate cancer (and so the arrows are labelled ‘s’).
DIAGNOSTIC HISTOPATHOLOGY 14:3
129
© 2008 Elsevier Ltd. All rights reserved.
Mini-symposium: Pathology of the Prostate
S
S
locally advanced prostate cancer
S
androgen ablation or anti-androgen therapy radiotherapy
S chemotherapy S S undetected prostate cancer S
new cases of prostate cancer
detected prostate cancer
S
S S
metastatic prostate cancer S
O
O
O
S
death from S prostate cancer
S chemotherapy O
S S localised prostate cancer
androgen ablation or anti-androgen therapy
S
S S
radical prostatectomy radiotherapy active surveillance
Figure 15 This is Figure 14 with the addition of outcomes (shown in green). The only outcome in this diagram that is shown is death from prostate cancer. Obviously in a full model we would have to include death from other causes, death from treatment of prostate cancer and quality of life, but in this initial model we are examining the effects of increased detection of prostate cancer on deaths from prostate cancer so we omit the other elements for clarity. We have put arrows labelled ‘s’ on all the states where a treatment is not curative, e.g. androgen ablation therapy for locally advanced prostate cancer (although that therapy does prolong life). The long loop from undetected prostate cancer to death from prostate cancer is an important one that might be difficult to detect unless an autopsy was performed. We add three arrows labelled ‘o’ from the three treatments for localised prostate cancer to metastatic prostate cancer because we know that in many cases those treatments are curative.
causal loop diagrams – balancing and reinforcing. A reinforcing loop will be more familiar as a positive feedback loop where a small change in one element can precipitate a sequence of events with large changes in many other elements. Reinforcing loops will not appear very often, if at all, in systems modelling human cancer but would appear with more frequency in models of infectious disease in humans when they would be the mechanism that would generate an epidemic. We can now add the different stages of detected cases of prostate cancer, their possible treatments and the possible movements between stages (Figure 14). These are all just straightforward positive flows of cases through a branching network and there are no loops within it.
DIAGNOSTIC HISTOPATHOLOGY 14:3
It is only when we add outcomes from the treatment that the diagram becomes more interesting (Figure 15). We know that there is currently no curative treatment for metastatic prostate cancer (androgen ablation therapy and chemotherapy prolong survival but are not curative), so any patient with metastatic prostate cancer is quite likely to die from that disease. We know that androgen ablation/anti-androgen therapy is effective for androgen-sensitive tumours but that tumours eventually become androgen insensitive so patients treated by this alone are likely to die of the disease. Undetected cases of prostate cancer in the population may cause death from prostate cancer that is only detected at autopsy so we can put a long arrow from that side of the diagram to death from prostate cancer. All these 130
© 2008 Elsevier Ltd. All rights reserved.
Mini-symposium: Pathology of the Prostate
S androgen ablation or anti-androgen therapy serum PSA screening S
locally advanced prostate cancer
S
radiotherapy
S chemotherapy S S undetected prostate cancer S
new cases of prostate cancer
S
detected prostate cancer
S
S S
metastatic prostate cancer
S
O
O
S
O
O
S
death from S prostate cancer
chemotherapy
radical prostatectomy
S S localised prostate cancer
androgen ablation or anti-androgen therapy
S
S S
radiotherapy active surveillance
Figure 16 We have taken Figure 15 and added the element of prostate screening by serum prostate specific antigen (PSA), which is linked to identified prostate cancer by an arrow labelled ‘s’ because we know it increases the detection rate of prostate cancer. We have then highlighted in red the pathways that would reduce deaths from prostate cancer that could be increased by prostate cancer screening – i.e. detection of localised prostate cancer that is then treated, thereby reducing the number of cases of metastatic prostate cancer. This is just to illustrate what would need to happen for PSA screening to reduce deaths from prostate cancer. The diagram as a whole shows that the number of cases in all the other stages would also be increased – a relative increase in detection of cases of localised prostate cancer would be needed to make PSA screening effective. The diagram also highlights the sorts of problematic cases that could be thrown up by screening – e.g. an asymptomatic man who is discovered to have metastatic prostate cancer.
arrows are labelled ‘s’ because they will increase the number of deaths from prostate cancer. On the benefit side of the equation we know that radical prostatectomy performed on localised prostate cancer is very often curative and so will reduce the number of cases of metastatic prostate cancer and subsequent deaths from prostate cancer. Radiotherapy on localised prostate cancer has a similar effect. Active surveillance of localised prostate cancer should also reduce the cases of metastatic prostate cancer although, to be totally accurate in the diagram, there will be some cases that appear to be progressing, which will require intervention by radical prostatectomy or radiotherapy to produce this effect. Now that we have created a simple model of prostate cancer and its treatment we can look at the effect of an intervention such as prostate cancer screening by serum PSA (Figure 16).
DIAGNOSTIC HISTOPATHOLOGY 14:3
The only pathway that will reduce the number of deaths from prostate cancer in the population is shown in red. The screening will cause more cases of prostate cancer to be detected in the population and, of these, a proportion will be localised that can be treated to prevent the development of metastatic prostate cancer with a subsequent reduction in deaths from prostate cancer. However, we can also see that the process of screening will also increase (at least on an initial screen) the number of detected cases with locally advanced and metastatic prostate cancer and it could be a challenge to decide how to manage these cases. If a man presented to his general practitioner with severe backache and was found to have bony metastases from prostate cancer, then the decision to treat him with androgen ablation/anti-androgen therapy would be clear-cut because he had symptoms from the disease that were reducing his quality of life. If a man was 131
© 2008 Elsevier Ltd. All rights reserved.
Mini-symposium: Pathology of the Prostate
found to have bony metastases from prostate cancer at screening but did not have any symptoms, then the decision to treat would be more difficult: we know that androgen ablation/anti-androgen therapy has only a limited period of effectiveness before the tumour becomes androgen insensitive so it might be prudent to wait until the man has developed symptoms before starting this treatment. However, the screening process has produced a man who thought he was well but has now found that he has cancer, which has spread through his body but the doctors who detected it don’t want to treat it at the moment – a difficult situation to explain to many patients!14 By looking at this simple model and following the different pathways that it presents, many such scenarios can be developed that might not have been apparent before the model was constructed. All these scenarios can be considered in a qualitative and semi-quantitative way before any decisions about whether or not to introduce prostate screening are made. The model can also be used as a basis for discussion between all parties involved in such decisions because it presents the information in such a clear and holistic manner. Any particular sector of the healthcare system can look at the model and assess the effects that any intervention will have on their work. Histopathologists could look at screening by PSA as an intervention and could predict that it would increase the number of prostate core biopsies and radical prostatectomy specimens that would be processed in their laboratories. They could also predict that there was unlikely to be any reduction in their workload from changes in other parts of the system because biopsies are rarely taken from metastatic prostate cancer. All of these analyses and predictions can be made from a simple model that would only take a few hours at most to generate from knowledge that healthcare staff already have. It might not be a perfect model, and different healthcare groups might derive different models for the same situation, but it will produce informed and holistic discussion about any interventions in the system that is likely to improve the quality of any subsequent decision-making.
using values of elements in their mid-range but perform badly at the extremes of these ranges (which still need to be tested, e.g. what is the value of prostate cancer screening if its incidence was 10 times less than its current level?). There are some methods that can speed up the development of quantitative models and produce reasonable estimates of values where there are several unknowns. There are some proprietary software programmes that enable causal loop diagrams to be encoded using a simple interface, and quantitative values to be entered, which will at least produce reasonable predictions of trends.15,16 These are valuable when a system becomes too complex to make predictions from simple visual inspection.
Conclusions A systems approach to prostate cancer can produce simple models that give insight into many aspects of its pathogenesis, diagnosis and management. The very act of creation of these models makes the creators focus on the available information in an analytical and holistic way, which is likely to enhance their understanding of the situation. Any models that they produce will be useful for discussing the situation with other stakeholders and could save a huge amount of time and, possibly, misdirected resources. The optimal level of payoff for model development in most medical environments is probably the causal loop diagram, which takes relatively little time to create but can be used to generate many scenarios and predictions. Although fully quantitative models are intrinsically attractive to many scientifically-minded healthcare staff, the resources that are required to develop and validate them and the intrinsic problems of making quantitative predictions from incomplete data make them less immediately useful. Nevertheless, such models remain a great challenge for the future. ◆
References 1 Sherwood D. Seeing the forest for the trees – a manager’s guide to applying systems thinking. London: Nicholas Brealey, 2002. 2 Cavana RY, Mares ED. Integrating critical thinking and systems thinking: from premises to causal loops. Syst Dyn Rev 2004; 20: 223–35. 3 Maani KE, Maharaj V. Links between systems thinking and complex decision making. Syst Dyn Rev 2004; 20: 21–48. 4 Wolstenholme E. Using generic system archetypes to support thinking and modelling. Syst Dyn Rev 2004; 20: 341–56. 5 Brawer MK. Prostatic intraepithelial neoplasia: an overview. Rev Urol 2005; 7: S11–18. 6 Iczkowski KA. Current prostate biopsy interpretation – criteria for cancer, atypical small acinar proliferation, high-grade prostatic intraepithelial neoplasia, and use of immunostains. Arch Pathol Lab Med 2006; 130: 835–43. 7 Ayala AG, Ro JY. Prostatic intraepithelial neoplasia – recent advances. Arch Pathol Lab Med 2007; 131: 1257–66. 8 Bostwick DG, Meiers I. Atypical small acinar proliferation in the prostate. Arch Pathol Lab Med 2006; 130: 952–7. 9 Richter E, Srivastava S, Dobi A. Androgen receptor and prostate cancer. Prostate Cancer Prostatic Dis 2007; 10: 114–8. 10 Grossmann ME, Huang H, Tindall DJ. Androgen receptor signaling in androgen-refractory prostate cancer. J Natl Cancer Inst 2001; 93: 1687–97.
Phase V modelling – quantitative models A quantitative model is often viewed as the Holy Grail of systems modelling. If the model in Figure 16 could be turned into a model where numbers could be entered into the various elements, then the effects of interventions could be quantified, e.g. the number of localised prostate cancers that would be detected from one round of PSA prostate cancer screening in men over 50 years in the UK. It is possible to produce such quantitative models but the required resources are orders of magnitude more than that needed to create a sophisticated causal loop diagram. The data for all the elements must to be obtained from reliable sources, usually from meta-analyses of randomised controlled clinical trials. The dynamics of each interaction in the system needs to be clearly and accurately defined and then encoded in an appropriate mathematical formula. A project that would produce a gold standard quantitative model of Figure 16 would require many months of work. There are other problems with quantitative models – any small errors in one part of the system are likely to be magnified (especially by reinforcing loops) in other parts of the system and may produce completely misleading results. Conventional quantitative models often given reasonable results
DIAGNOSTIC HISTOPATHOLOGY 14:3
132
© 2008 Elsevier Ltd. All rights reserved.
Mini-symposium: Pathology of the Prostate
11 Thompson IM, Ankerst DP. Prostate-specific antigen in the early detection of prostate cancer. CMAJ 2007; 176: 1853–8. 12 Fitzsimons NJ, Sun L, Moul JW. Medical technologies for the diagnosis of prostate cancer. Expert Rev Med Devices 2007; 4: 227–39. 13 Ilic D, O’Connor D, Green S, Wilt T. Screening for prostate cancer: a Cochrane systematic review. Cancer Causes Control 2007; 18: 279–85. 14 Rimer BK, Briss PA, Zeller PK, Chan ECY, Woolf SH. Informed decision making: what is its role in cancer screening? Cancer 2004; 101(suppl 5): 1214–28. 15 iThink, Stella. isee systems inc., Lebanon, New Hamphire, USA. Available from: http://www.iseesystems.com/ [accessed 26.02.08] [iThink and Stella are the same software product but the literature that accompanies them is slanted to business applications (in iThink) and education/research (in Stella)]. 16 Vensim. Ventana Systems, Inc., Harvard, Massachusetts, USA. Available from: http://www.vensim.com/ [accessed 26.02.08].
DIAGNOSTIC HISTOPATHOLOGY 14:3
Practice points • A systems modelling approach provides useful insights into the pathogenesis, diagnosis and management of prostate cancer • State and influence diagrams provide a clear mapping of current situations that can be useful for explaining diagnostic and treatment protocols • Causal loop diagrams are very useful for making qualitative and semi-quantitative predictions about any changes in a system of diagnosis and/or management of prostate cancer • Quantitative hard systems models take a huge amount of resource to develop and validate, and should only be used when softer methodologies have been fully explored
133
© 2008 Elsevier Ltd. All rights reserved.