A Guide to Implementing the Theory of Constraints (TOC)





Next Step



Bottom Line


Supply Chain

Tool Box



& More ...



Lead Times

Finished Goods


& Distribution

& Marshalling

& Healthcare



Patient Waiting Lists & Healthcare

In the previous page we saw how to apply replenishment to log marshalling.  Of course people are not logs – we know that.  People are much more perishable, especially when they are unwell.  However, is there something that we can learn from log marshalling in general that can also be applied to healthcare?  Let’s run a test, let’s compare the patient waiting/referral process against log marshalling and see if it is different or similar.  If it is similar then maybe we already have a reference environment from which we can extract the principles and apply to healthcare.  Let’s see.

Firstly, however, if you have arrived at this page directly rather than sequentially through the replenishment page and the distribution page, then please consider reading the replenishment page first.  This will ensure your understanding of the technical solution (the planning and control system) that we are going to apply in this case.  Forearmed with such knowledge you will be in a much better position to evaluate the description of the current problem and also the potential for the detail of the solution.

The “producers” or the source node in this system – the general practitioner, or local doctor – is in this case a service operation having no set-up and appears to produce stock for the system – patients – at a random rate and in units of one.  Now, however, there is no longer a geographic many-to-one relationship between producer and the next node, the specialist.  But rather there is a many to many relationship.  Any general practitioner may wish to refer a patient to any one of a number of different specialists depending upon the nature of the illness.

Let’s try to draw this network.

All we have done here is to have changed the relationship from one-to-one in log marshalling to many-to-one in public healthcare and changed the labels.  As a generalization then, the marshalling network seems to describe the flow of patients.  There is however one interesting and critical difference in this network, the actual production part isn’t at the source nodes any more, here labeled the local doctor.  The actual production is at the end node, the intervention/admission.

Health professionals “bristle” when manufacturing analogies are used in health, “we don’t make cans of beans you know!”  However, we need to use manufacturing terminology for a moment to describe part of this system in consistent terms.  When the intervention is carried out in a theatre, that part of the process is a production process.  It has a set-up, it has particular equipment and staff for particular procedures, and it may even have particular rooms.  We need to know this to distinguish it from the supply chain portion, the referral and waiting list part, because the way in which the supply chain portion and production parts are managed are intrinsically different.

How can we be sure that intervention is a production process and not a project process?  Well, if we look at an operating list we can characterize it in terms of patients per day, rather than days per patient.  Therefore we can be quite certain that from the system’s perspective the intervention node is a production process.  This is important because it is likely that the intervention node will always be the control point – the drum in drum-buffer-rope terminology – regardless of whether the constraint is internal (we don’t have enough capacity) or external (we don’t have enough patients).

How can we be sure that this node will always be the control point?  Well, at a guess, it is the most capital extensive step, either a new theater in surgery or a new ward in medical cases.  It certainly will be the operationally most expensive step in terms of on-going staffing and support.  Therefore it is unlikely that new theaters will be built in rapid response to demand.  The control point is therefore unlikely to shift somewhere else as a consequence of additional theatres being built.

So it looks as though there is some validity in investigating marshalling or a convergent supply chain as a descriptor for patient waiting lists.  However, this brings us to a significant issue – waiting.  On the very first page, the introduction to these webpages, we mentioned that we can batch in time or we can batch in quantity.  Supply chains are dominated by batching in time, and the patient waiting list is a supply chain.  Batching in time is so pervasive in healthcare that we absolutely accept it as normal.  We fail to even question the reasons for its existence.  Whenever we batch in time we cause waiting to occur.  Reducing batching in time, along with removing the policy constraints that limit output; will substantially reduced patient waiting lists.  Would this be a worthwhile cause to pursue?  I think so.  Are you interested?

Good, then we need a plan of attack.

Plan Of Attack

There is only one plan of attack, the 5 step focusing process that we have used to date in the analysis of all of our logistical endeavors.  Let’s repeat it here for good measure;

(1)  Identify the system’s constraints.

(2)  Decide how to Exploit the system’s constraints.

(3)  Subordinate everything else to the above decisions.

(4)  Elevate the system’s constraints.

(5)  If in the previous steps a constraint has been broken Go back to step 1, but do not allow inertia to cause a system constraint.  In other words; Don’t Stop.

What is the constraint in this system?  What are we trying to protect?  The constraint isn’t a lack of “customers” in this instance; there is no shortage of patients.  It something else; it is expensive and finite capacity – limited physical space somewhere, possibly “funded beds” (as opposed to beds that exist but are deemed to be unfunded) for medical conditions or theatre space for surgical conditions.  That almost answers the second question – how to exploit the constraint.  We lose output from the system whenever we have patients who need intervention, but who are not in the right place at the right time to receive that intervention.  Healthcare is a service; we can’t store the intervention for use at a later time.

In order to exploit the system we need to ensure that we can never waste an opportunity to carry out an intervention.  We will need to deduce how to best do this – a strategy for exploitation.  We will also need to develop how to best subordinate the rest of the system once the exploitation strategy is in place.

To properly and successfully exploit and subordinate the valuable capacity in this system would mean that we begin to have shorter waiting times and eventually spare capacity.  First however in order to be able to determine the exploitation and subordination strategy we need to examine the properties of waiting list networks in a little more detail.  Let’s do that.

The Waiting List Network

Let’s acknowledge right at the outset that there are two ways that we could effectively exploit the constraint – one way is to decrease the input and the other is to increase the output.  We could express this as follows;

(1)  Approaches that avoid illness in the first place.

(2)  Approaches that mitigate or cure illness once contracted.

Although there is now an increased emphasis on pro-active prevention rather than reactive intervention, the fact remains that much illness still requires active intervention and moreover there is a backlog of work.  This backlog arises from rising general expectations from taxpayers and improved levels of care/technology from practitioners.  Technology is a double edge sword here, it both substantially reduces the effort in simple interventions freeing up bed space that was unimaginable 20 or 40 years ago, and at the same time making possible interventions that tie up bed space that was unimaginable 20 or 40 years ago.

We need to recognize that the patient waiting list network in the main is concerned with non-acute admissions although how we handle this network strongly impinges upon acute work also.  Public health systems must deal with non-acute, acute, and emergency patients; all at the same time.  But this is not unusual.  I have not yet seen a process that didn’t operate a concurrent but differential priority system of some sort.  Concurrent differential priorities are the rule not the exception in serial processes and health professionals need to recognize this commonality.  In any system; manufacturing, service, or supply chain, the only way to manage differential priorities concurrently is to have adequate sprint capacity and/or buffering.

Let’s examine our non-acute network then, from the perspective of a patient/taxpayer (tired pun intended).  It is a bit radical to take a patient’s perspective but let’s press on.  The patient books an appointment with the local general practitioner who determines that the problem more correctly needs specialist assessment.  The doctor might ask “public or private” implying some differential service but we won’t go there.  Our patient paid taxes damn it and is going to go public.  Well, please wait 2 weeks and you will get a letter from the hospital telling you when to attend an outpatient’s clinic.  Two weeks pass and the letter arrives – please attend a clinic in 6 weeks time!  Six weeks pass and the specialist appointment date arrives.  And isn’t it funny how all the other people in the clinic seemed to have similar conditions.  Anyway an assessment is made and intervention is recommended within 6 months.  Ah, that doesn’t mean 3 or 4 months, that means something like 5½ months or 6 months.  That’s 6 months waiting plus 1½ months waiting plus ½ a months waiting.  That’s 8 months waiting all up.  That is, if the specialist didn’t refer you back to the general practitioner.

Why all the waiting?

Well, we are just trying to be efficient didn’t you know!

If you look at the marshalling system it looks like an “A” – upside down I grant you.  However A-plants describe a situation where there is general convergence in manufacturing.  Patients’ waiting lists don’t manufacture anything, but as a supply chain they seem to exhibit the same behavior as A-plants.  Let’s look at an A-plant description; under traditional management practices in A-plants the tendency is to misallocate resource time in an attempt to maximize efficiency and utilization figures.  Large batches are used to keep the measurements high resulting in a poor component mix and constant shortage of the right parts (1).  Furthermore these large batches move in waves throughout the plant causing temporary bottlenecks to wander from resource to resource.  Since material is constantly out of balance, overtime is used to ‘catch up’ so that shipments can be made on time (1).  We can expect similar things to happen in patient waiting lists.

What evidence do we have of local efficiency measures to substantiate this assertion?  Well just reflect for a moment on the number of measures that we must compile for reporting and how many of these directly relate to improving wellness in the community.  Aren’t most really local efficiency figures?  Aren’t most of them sheer frustration to substantiate even on a good day?

What, then, if we were to increase the frequency of clinics in such as system – and indeed interventions as well.  Nothing major, moving things from once a fortnight to once a week or from once a week to twice a week.  What effect would that have if we could simultaneously address the backlog?  Think about it.

But Demand Will Increase!

Well, there is a very peculiar notion in healthcare that if we improve patient service levels then demand will also increase.  We need to examine this.  This notion belongs on another planet.

Would you ever wish upon yourself a serious illness?  Pretty damn unlikely.  So if patient waiting lists decrease, and service improves, are people going to become ill more often just to avail themselves to the new levels of service?  Pretty damn unlikely also.  So where does this curious notion arise from?

One situation where it might arise from is if we are currently failing to meet a real need (as opposed to a desire, or a want, or what the marketers would call a latent need – one when you go out and buy something you didn’t even know that you wanted).  If we are failing to meet a real need and we increase availability then of course there will be an apparent rise in demand.  However, that demand was already present, it simply wasn’t being met.  Failure to meet a present and real demand is not a reason to restrict services if it is possible to improve access those services using existing resources.

There is another aspect to this apparent paradox.  Healthcare has very strong negative reinforcing loops operating in it.  Failure to respond to a need at an early stage means that the need when eventually met consumes far more resources.  This might help explain why demand is perceived to increase.  It is not necessarily the total incidence that is increasing but rather the severity of the individual incidences once they reach an agreed level for intervention.  We will return to this thought later.

You Don’t Understand We Have Too Many Acute Patients

Yes but, we have too many acute patients.  This is an interesting problem.  Some people become acutely ill suddenly.  Others become acutely ill over time – time spent waiting for non-acute intervention.  Governments find it hard not to fund acute work and not so hard not to fund non-acute work, so we can guess why acute work load is, in part, so significant.  Moreover, we already know how to break this vicious cycle – earlier intervention.  Too many acute patients is really just an excuse.  There are others too.

The local population is too old; the local population is too young.  We have diverse socio-economic challenges in our area, the area is too rural and dispersed, the area is too urban and condensed.  We can’t retain good staff, we can’t attract talented doctors to our specialty, our specialty is under-recognized, our staff are older and more expensive than the mean and so forth.

What about; our young doctors are attracted to the large cities, or (if you are in a large city) our young doctors are attracted overseas.  And you can’t get good locums anymore.  Our buildings are over 40 years old, our buildings are an earthquake risk, the air conditioning is antiquated, our total corridor length is much greater than anywhere else (its true trust me).  The flu season was early/late this year – but never on time.  We have an unreasonable orthopedic load, we are a national center for ……… but this isn’t recognized in the funding.

In fact, you have probably worked out that there is no end to this list.  But don’t worry these are not the problems either.  They may be symptoms of people’s frustrations, but they are not the core problem, and therefore solving them is not the solution (but that has never stopped anyone yet).

Let’s move on.

Cart Before The Horse

Improving the patient waiting list network is indeed a fine ideal, especially if it ensures that we don’t miss an opportunity to do an intervention because we didn’t have a patient ready – even though there are 100’s of patients on the list (it happens!).  However, the reason supply chain follows production on these webpages is that we have to sort out our production first if we are to improve our supply chain.  Now there are exceptions to this; for instance where we don’t own the production stage or it is beyond our span of control or sphere of influence.  In these instances then, yes, indeed we have to do our best in spite of the limitations. However, this is not the case in public health systems, the production stage and the supply chain stage are integral.  So we should address the production side first.  And if we can’t win there, then that shouldn’t be an excuse not to look at the supply chain mechanics nonetheless.

In order to increase the production side we must address a policy constraint.  Let’s have a look then at that.

There Is No Goal In Public Health

It has been said that if there is no goal then the absence of a goal is the constraint.  We also noted in the measurements section that the goal of a system is in fact open-ended.  You can’t have enough of the goal.  In contrast the necessary conditions that support the goal can be viewed as having limits.  Once we satisfy a necessary condition additional satisfaction does not improve the rate at which the organization moves towards its goal.

A problem arises however in not-for-profit organizations of which a public health service is just one example.  Scheinkopf notes that in not-for-profit organizations “there is a tendency to believe that the measures are so intangible and that attainment of purpose is such a subjective call, that such measures are simply not discussed.  The focus ends up to be on measuring and managing the things we call ‘tangible,’ such as money (2).”

New Zealand health boards must currently meet an 11% “capital charge” on some types of new investment (the rate-of-return incidentally is one that some public companies in the free market can currently only dream of).  This means that the Government must pay the health boards pro rata 11% too much to cover this capital charge, which the boards then pay back to the Government showing that they are indeed efficient.  This financial efficiency can be met by restricting access by raising the level (points) for non-acute admissions.  At best, meeting the capital charge is a necessary condition and a perverse one at that.

Because the capital charge is imposed upon the system from outside and must be met, it is a necessary condition for success.  However, necessary condition aside, it is just a Government policy, this doesn’t mean that its validity shouldn’t be challenged, nor the cost mentality behind it.  However the very real danger is that once the necessary condition has been satisfied (boards run a “balanced” budget) there is no driver for further improvement.  The necessary condition, due to its prominence, is mistaken as a goal – which emphatically it is not.

It is no exaggeration to say that public health is missing a goal.  Instead it has, as an objective, a necessary condition – meet budget.  We can illustrate this further.

“The Nelson-Marlborough District Health Board confirmed all elective surgery will be postponed for about six weeks over summer.

The moves come at a time when some patients are waiting up to five years for non-urgent surgery, and the board is preparing to cut people from its waiting lists if their conditions are not considered serious enough to warrant treatment in the public health system.”

“The Health Ministry contracted the board to do fewer operations than it had the capacity to perform.  As a result it was already significantly over budget less than four months into the financial year.

The purpose of the cuts was to reduce surgery to contracted levels and save money (3).”

If you believe in reductionist/local optima viewpoint you will also believe that each operation has a cost and by avoiding operations we can avoid all the costs associated with them and thereby save money.  If you understand the systemic/global optimum viewpoint then you know that such efforts will hardly save a penny.  Sure it will save on some variable costs.  However, using the quote above as an example, we should ask what will the buildings do for 6 weeks, what will the staff do for 6 weeks, and what will the air conditioners do for 6 weeks?  They are not variable expenses.  And of course what is the final cost to the system when the work is finally undertaken – is it more or is it less?

Let’s have a look then at evidence of, not of postponement, but of removal from a list.

“Many gallstone patients in Auckland must now suffer at least four attacks of severe pain and vomiting in a year to qualify for surgery.

Or they must have two attacks of gall bladder inflammation, or experience worse symptoms or complications.

Less that four pain and vomiting episodes, called biliary colic, and you would probably fall below the cut-off point – set in response to Government funding levels – for elective surgery at North Shore Hospital (4).”

In the same article but a different hospital.

“Waitemata officials started to introduce their new scheme in November after struggling, like all district health boards, with having more patients than can be treated.  They hope to extend it to other types of surgery later.

Under it, about 40 patients have already been taken off the surgery waiting list because they are not considered sick enough.”

Note that the qualifier is 4 attacks in one year – and then you go onto an “elective” waiting list; but there was no mention of how long the list is until intervention.  Removing people from waiting lists who are deemed not ill enough to warrant treatment conforms exactly to one of Senge’s system thinking archetypes – eroding goals (5).

So I think it is safe to say that the objective illustrated here is characterized by a limiting necessary  condition and that necessary condition is to meet budget.  We don’t have a goal at present.

Who Should Set The Goal Then?

So if we don’t have a goal at present in the health system, who should set the goal then?  Well, the answer is clear, the owner of the system should set the goal.  And the owners are the taxpayers aren’t they?  Sure, but the Government of the day administers the health service on behalf of the taxpayers; so it is the Government who is the proxy owner of the system in this instance, and it is the Government that should set the goal.

The Government currently sets a number of necessary conditions that are financial in nature because articulating a non-financial goal and the fundamental measures to support it is deemed to be too difficult.  But is it really that difficult?  Let’s try.

How Do We Set The Goal?

Let’s try and set a goal for public health so that we can move along back to our objective of showing marshalling as a viable model for public waiting lists.  How do we do that?  How do we set the goal?  I guess that we need to ask where we want the public health service to be at the present.  That would be a good place to start.

A politically correct goal might then become; a timely and appropriate outcome.  But what is the outcome?  Is it community wellness?  If it is community wellness, are we seeking to maximize it?  That certainly seems open ended as a goal should be.  However, it might also imply incorrectly that funding should be maximized and clearly there is a problem here because most people don’t want taxes to increase which is exactly where the funding must come from.

Then, how about; improve community wellness, as an appropriate outcome?  Improving community wellness seems sufficiently open-ended at this point in time (maybe even bottomless), we could certainly do with a lot, lot, more of it.  Why don’t we run with this for a while and see if it will work for us.  Thus the trial goal for a public health system is to; improve community wellness now and in the future.  Let’s write that.

Establishing a goal is fine; however, we now need to ask what are the absolute necessary conditions or inputs that will give rise to this goal.  In order to obtain this goal it seems that there are at least 2 necessary conditions that we must satisfy.  We alluded to these in defining the goal.  A timely and appropriate outcome implies a timely and appropriate input.  The appropriate input could be pro-active prevention or the reactive intervention that we carry out.  The timeliness depends more upon availability at this moment than anything else.  So let’s add these two necessary conditions to our goal.

It seems that the appropriateness of the intervention isn’t so much in contention as the timeliness.  It seems then that one necessary condition is currently satisfied – the appropriateness.  Medical professionals do not appear averse to taking up new approaches or technologies in either treatment or prevention.  However, the other necessary condition – timeliness, isn’t currently satisfied.

In fact, satisfying this non-financial necessary condition looks a little untenable.  The proverbial rock and a hard place.  We need to improve the outcome – community wellness – with a level of availability and therefore timeliness that many would consider is currently “insufficient.”  It therefore would be too easy to write another necessary condition leading into the current one saying “secure sufficient funding” in order to increase the level of availability and therefore increase the timeliness – however, it would be quite another thing to actually receive that funding.

We should also remember from the measurements page that a not-for-profit organization such as a public health service must watch its operating expenditure against its existing fixed level of funding least it runs a deficit (6).  So running in the red and hoping is out as well.  How then do we ensure sufficient timeliness and maintain our operating expenditure at the same time?

Let’s go back to one of the most important statements in Theory of Constraints;

Productivity = Throughput / Operating Expense

We have muddied the water a little because our goal is now non-financial and throughput, as defined, is a financial measure (sales – totally variable costs excluding direct labor).  However, we can jury-rig another equation that will do just about as well – we will substitute output for throughput;

Productivity = Output / Operating Expense

We can measure our output – patients.  We can measure our input – operating expense.  If output goes up and input goes down or stays the same then we have increased our productivity and we have also moved towards our goal.

Let’s be clear however, increased productivity doesn’t mean working harder.  It does mean though, knowing sufficient about the system, its dependencies, and the variability in and between dependencies that we can protect the most valuable or most important part, that part that we have the least capacity to spare.  Let’s make “not working harder” an explicit necessary condition to our goal so that this aspect is not misinterpreted or misrepresented.  Let’s draw it.

This then is the goal and the necessary conditions for a public health system.  We have identified a non-financial necessary condition – timeliness – that is currently not being satisfied.

How Then Do We Measure Progress Towards The Goal?

Using productivity as a measure of progress towards the goal is a bit of a blunt weapon – in fact it is more an indication of the method than the measurement that we should use.  The fundamental measurement then is our non-financial necessary condition, the one that we are failing to meet currently – timeliness.

The Government – the owner of the system – must set maximum national patient wait-times that must be met.  We can measure this performance and it is non-financial.  Moreover we can see that meeting an increase in demand at static maximum wait criteria and funding must mean an increase in productivity.  Also meeting a lowered maximum wait criteria at static demand and funding means an increase in productivity.  We can measure progress towards or away from the fundamental measurement with two local measures; patient-days-wait, and patient-days-late.

Yes But, The Government Already Uses Maximum Patient Wait Times

So how is our proposal different then; the Government already uses maximum patient wait times for many aspects of healthcare?  That is true, but how is the issue managed at present?  Timeliness is currently managed not by increasing productivity but by decreasing productivity.  It is managed by raising the criteria for consideration, so that the patient wait times may remain high and constant but the level of unwellness in the waiting list becomes greater over time and the number of people treated becomes fewer and fewer – we saw direct evidence of this in the earlier quotes.

Moreover, the maximum wait times are measured purely by the number of patients.  Our local measures; patient-days-wait and patient-days-late are much more revealing about the true nature of the waiting list.  But we ourselves must wait a little before we can investigate this aspect further.

Broader Issues

First, however, there is a broader aspect to productivity that applies to a public health system.  Public health systems are not “stand-alone,” public health productivity impinges upon the productivity of the whole nation/state.  Consider for instance a country with first class productivity in one of the primary industries such as; agriculture, fisheries, forestry, or mining, or first class productivity in any one of the secondary manufacturing industries.  These activities generate national income.  Why do we constantly strive to increase the effectiveness of these national income generating activities if a major consumer of this income, healthcare, operates on assumptions once thought valid in a previous century – and I mean the 19th not the 20th century.  Other sections of the economy have moved on.

Currently most hospitals are implementing some form of patient information management system and some form of enterprise-wide scheduling system.  Enterprise-wide scheduling systems were described in the section on production, essentially they are finite scheduling solutions based upon a reductionist/local optima approach.  As we know from manufacturing, reliance on these techniques depends on excellent data integrity but generally results in increased work-in-process because they fail to protect the system from variation even through they have ample protection embedded within the schedule – in short they fail to protect the constraint – output goes down, work-in-process increases.  Increased work-in-process in this environment means more patients-in-waiting and waiting for longer.

The reality is that in both manual and automated scheduling systems many theater opportunities are lost due to poor protection of the constraint.  These losses are buried in the general theater utilization hours, we have to scratch the surface to find them, but they are real, and they do present a real opportunity to improve output at current operating expense.  And that brings us to our critical erroneous assumption.

A Critical Erroneous Assumption

There is an assumption that we totally fail to challenge – the assumption that we are sufficiently productive and that we can not improve further.  The pervasiveness of this assumption can be demonstrated every time someone says; ”yes we could process more patients if only we had access to more funding.”  The hospital in the earlier quotation is very likely to have sufficient productivity – it could do more operations than contracted for (don’t be fooled by contracted cost, you need to see the flow of money in and out of the system).  The real issue is that if one hospital provides a better level of service than others it is defying a charter that requires equitable access to all people in all parts of the country.  This means other hospitals are currently not as productive.  The most productive hospital and all other hospitals in between must be hobbled to the level of the least productive hospital in the system in order to ensure equitable access.  Think about it.

Yes but, all the other hospitals could improve to the same higher level couldn’t they?  Well you would think so; this would be the ideal situation.  However, there are two reasons why this doesn’t happen.  Firstly under a reductionist/local optima costing process, if we improve our productivity our unit costs will go down and next funding round we will receive less to do the same number of procedures rather than the same amount to do more.  This is a very real fear of hospital management.

The other reason is more important.  Currently in the health system there is little knowledge of the rules of engagement that we first saw in the measurements section.  Let’s repeat them here;

(1)  Define the system.

(2)  Define the goal of the system.

(3)  Define the necessary conditions.

(4)  Define the fundamental measurements.

(5)  Define the role of the constraints.

As you can see, in health at present we have just a few financial-based necessary conditions; we are missing so much of the whole picture.  Why won’t we do this if it is so simple?  Are we scared?  No, I don’t think so.  It might be that many people simply don’t know how to evaluate the role of the constraints in this system yet – or that they do know how to but common practice runs counter to this.

Well, fortunately we are using our common sense rather than common practice, so let’s continue with our examination of patient waiting lists and marshalling.  We really ought to stop looking at the problem and start looking at the solution.


How can we describe the actions of the nodes in the patient waiting list network?  We have suggested that the supply chain here is a marshalling supply chain, or more accurately marshalling and consolidation.  Patients are marshaled in by referral from numerous local doctors and consolidated into specialties and then lists.  The consolidation is carried out in accordance to a push-to-need basis.

General practitioners feel that a particular patient needs specialist expertise (and it is the expertise of the general practitioner to know when this is required) and “launches” the patient into the process and hopes that the outcome will be favorable (and timely).  As in all other supply chain solutions here we need to replace this with some sort of pull-and-replace system.  The constraint, the most valuable and limited resource, must pull the patients via the waiting list network to a position where they are ready to receive intervention as soon as possible.  Maybe we should call this a pull-to-cure or a pull-and-cure system.

If at some future point in time there are insufficient patients to fully load the system then we are moving in the right direction.  And if currently we can at least stop waiting list expansion (without fiddling with the criteria) and affect a contraction then we know that we will eventually reach that future point.  The key is that the system must initially pull at a faster rate than the incidence rate of the problem.  How are we going to achieve that?

Well, unlike the distribution problem or the log marshalling problem, where the constraints in the system were non-production constraints, here the constraint is a production constraint.  The intervention produces something; it produces favorable outcomes – but currently it produces an insufficient number of them.  Thus we need to break our solution into two subsystems;

(1)  Production subsystem – Intervention.

(2)  Supply chain subsystem – Patient Waiting List Network.

And as you know, common sense tells us that the answers are already in the system.  So let’s have a look.

General Solution – Part One; Intervention & Drum-Buffer-Rope

A solution with an unusual name and very powerful consequences.  Drum-buffer-rope is the Theory of Constraints production solution, it is a logistical solution.  It is fully described in the section on production; it is really a way of thinking more than anything else – a way of thinking that enables substantially increased output from constrained situations without recourse to additional funding or manpower.  There is a good example from neurosurgery in the United Kingdom (7).

The Radcliffe Infirmary went from canceling 64 elective neuro-surgical procedures over a 3 month period to canceling none in the same period the next year.  Out-of-hours operations were drastically cut and output went up by 16%.  Would a reduction in non-acute cancellations be useful to you?  Would reduced out-of-hours operating be useful to you?  Would an increase in output be useful to you?  This is not a trivial solution.

We could get away here with just briefly mentioning some aspects of the drum and the buffer.  The drum is the constraint, it beats out the rate at which the system works at.  In our case the constraint is most probably a surgical theater or a medical bed.  Let’s draw this using our systemic model that we developed earlier.  The constraint – our drum – is the rate limiting step.

A buffer is quite tightly defined – in this situation it is a measure of time, the time for a patient from the moment of admission to the beginning of intervention.  To properly exploit our scarce resource in surgical cases we must admit patients in good time so that they are always ready for intervention.  However, to properly subordinate the scarce resource we must also not admit too many patients at any one time.

Watch the distinction; it is very, very, important.  After all, one of our local ward measures is average “bed nights” or some such similar measure.  Having a lot of patients for a short time is locally positive; having few patients at any one time for longer is locally negative.  The current local measures do not support the global objective of the system.  If we have too many patients waiting for too short a period we will absolutely miss some interventions “because the patient wasn’t ready”.  Hell, the patient was ready.  It was the system that wasn’t ready.  Our output goes down.

If we have fewer patients waiting longer between admission and intervention we won’t miss an intervention, output goes up.  System operating expense remains the same.  It seems counterintuitive, but if it was intuitive we would have done it – right?

We could summarize this as follows;

Introducing constraint buffers and decreasing process batch size automatically
 aligns the process with the goal

What do we mean by process batch?  Well, I guess that an operating list is a process batch.  The other sort of batch size that we might refer to is a transfer batch, and in a service operation like this a transfer batch will be in units of 1 – the patient.  To decrease the size of the process batch means that instead of operating all day only on Tuesdays for instance – and causing uneven ward work-load, how about Tuesday and Monday and Wednesday mornings instead.  Forget the detail, it is simply that we are trying to decrease the number of patients at any one point and increase the frequency.

Really we are trying to better balance the flow.  Again be careful, we never balance capacity but we always try to balance the flow – just the opposite from local optimization.  Of course there are practical limits to this, but we should make sure that the limits are real and not policy.  We need to make sure that the policy is not some assumption rooted in the 1960’s or the 1950’s.  Increasing the production frequency is the primary driver that flows on back up into the supply chain – the patient waiting list network.  We had better look at that next.

General Solution – Part Two; Patient Waiting Lists & Replenishment

The constraints in this system are in the intervention stage, the stage located within a hospital, and this is the stage that we must exploit.  Therefore, all other stages are non-constraints and we must subordinate these to the constraint.  The patient waiting list network, like the log marshalling network, must subordinate to the constraint until such time as there is a substantially reduced waiting list and additional patients present at admission at a rate that is less than rate of intervention.

To properly subordinate we must ensure that the waiting list network never “starves” the production node.  It starves the production node when it fails to produce a patient for admission in good time.  It happens.

Talk to a scheduling clerk and you will hear stories like “I need a patient for the operating list on Tuesday fortnight – and I have rung and rung around the patients on the waiting list but do you think that I can find one!”  It’s amazing, but true, and very frustrating for those trying to do their very best.  Thus our intuition as well as a good dose of common sense suggests that we should move patients through the waiting list network as quickly as possible to the place of greatest aggregate safety for both the patient and the system – just prior to admission.

In fact, in medical cases, it is likely that the supply chain prior to admission will also form part of the constraint buffer.  This type of situation is not so uncommon in manufacturing especially where the first step in the process is so capital-intensive that to “buy another one” is prohibitive.  The expense in both cases here is the bricks and mortar and the considerable number of skilled staff required to run the facilities around the clock.

Let’s consider some questions then;

What would happen if we could increase the frequency of clinics prior to acceptance for intervention?  As an example, instead of holding a clinic once a month for a day (because it is efficient for staff) what about holding a clinic for half a day every fortnight, or until mid-morning every week?  It is kind of like waiting for one 747 or one of two 737’s.  The total waiting time is less for the smaller more frequent service.

What potential could that have?

What about if we could remove nodes completely or combine nodes so that they occur at the same time and place, maybe carry out some tests on the same day in the same place for instance?  What potential could that have?

Hold on to these thoughts for a moment.

As in linear supply chain, distribution, and log marshalling, we need to introduce into this system the Theory of Constraints supply chain solution – replenishment.  If you are unfamiliar with fixed-frequency variable-quantity replenishment then please check the explanation on the replenishment page – it is important.

Each node in the waiting list network becomes a buffer for the next node containing sufficient patients to ensure that it can supply the next node down while it pulls patients from the next node up.  The constraint, the drum, in the production portion is the originator of this pull signal.  Let’s draw the supply chain portion then.

If we carry out replenishment correctly we will move safety to the area that is most important, the area closest to admission.  Let’s draw that.

And if we increase the frequency of the clinics and other waiting list processes then these buffers can be very small indeed and passage from one end of the list to the other will be very rapid.  That way once a referral is made the patient can move through the system quickly and be available to be “worked upon” – either an operation or a medical treatment if required as soon as possible.  We can summarize this;

Introducing replenishment buffers and increasing resupply frequency automatically
 aligns the process with the goal

Now if we return to those thoughts that you are holding on to, there is probably a big red flag saying’ “yes but there are too many patients-in-waiting in the system now to make such a process work.”  Yes there are.  But unless we get the appropriate mechanism in place even before it is apparently needed things simply can’t improve.  If we were to size our buffers today we would find that they are way over-full.  But at least we know where we are heading.

In every situation where the system is drowning in work-in-process, people are reluctant to give up the system that causes the work-in-process that drowns them; “because there is so much work in the system that doing this will have no effect.”  Exactly wrong.

We recognize how chaotic huge numbers of patients-in-waiting are because periodically we “fiddle” with the criteria to try and reduce the numbers.  Unfortunately that just feeds a negative reinforcing loop – we get more acute patients.  The only solution is to maintain the criteria and increase productivity.  You will be very, very surprised at the effects.  Patients are not logs, and they are not cars, but that doesn’t mean that we can dismiss the principles.  Well in fact we can dismiss them, but they won’t dismiss us.

We need to cut the strong negative reinforcing loops and replace them with strong positive reinforcing loops.  We need to look for systemic/global optimum solutions not reductionist/local optima solutions.  We need to look at trying to reframe the environment and not to continually applying band-aids.  The solutions are already in the system, and those solutions although they represent change, represent a change in meaning only.

Let’s Put It All Together

Let’s try and pull all of this together by showing the system in its proper order; the supply chain patient waiting list network feeding into the intervention stage.  Likely as not there is another supply chain at the other end – district nursing, but let’s leave that for another day.

Doing this it becomes clear that there is a feedback between the two.  We need to make sure that we don’t ever waste our scarce intervention stage, and at the same time we need to ensure that the supply chain doesn’t ever fail to provide an appropriate patient at an appropriate time.

Local Performance Measures

Earlier we parachuted in a goal for public service healthcare and looked at how to measure whether we are moving towards the goal or away from it.  The goal and necessary conditions might provide a measure for a whole system, but how do we know in a system as complicated as a large public hospital or a district health board that the parts – the subsystems – are also aligned and moving in the right direction?  Really we are asking; how do we know that the non-constraints are subordinated to the constraints.  For this we need local performance measures.

Another way of looking at local performance measurements is that they should judge the quality of the execution of the exploitation plan (8).  What is the plan in this case?  Surely it is to provide a timely and appropriate outcome.  We can’t comment here on the appropriateness but we certainly can on the timeliness.

Timeliness is reflected in two particular measures;

(1)  Unit-days-wait.

(2)  Unit-days-late.

In fact of the two, late-days is more important, but waiting-days always seems easier to explain first.  These two measures are simply a re-verbalization of the two measures that we have consistently applied to any subsystems in production or supply chain processes.  In fact, we used these exact measures to introduce the concept of local performance in the measurements section.  Let’s have a look at these again in detail.

Let’s say for instance that a certain outpatients’ clinic for referrals has 50 people on the waiting list at any one time and last year these people waited on average for 12 weeks, this year we still have 50 people on the waiting list at any one time but they now wait on average for 16 weeks.  What is the total waiting time here?

Well, we know that last year that there was on average 12 weeks by 5 days per week by 50 people = 3000 patient-days-wait.  In comparison, this year there are 4000 patient-days-wait on the list.  Is the performance better or worse?  It’s worse of course.  If we can stop patient-waiting-days from increasing, or better still reduce it, then we must have improved the system.  Let’s add this measurement to a linear representation of our health system (both patient waiting list network and hospital intervention).

So waiting-days is one measure that we can use to evaluate a subsystem with, or indeed even departments within a subsystem.

Another aspect of timeliness is that regardless of how long we must wait, do we still receive attention “in time” at the end of the wait or are we late?  Let’s continue with our analogy.  Let’s assume that last year our patients were expected to be seen by a specialist within a recommended guideline of 12 weeks of referral.  Some, however, weren’t seen within this time-frame.  Let’s say that 3 patients were seen after 13 weeks and 2 were seen after 14 weeks.  Again we might argue that just 5 out of 50 or 1 in 10 patients were not seen within the recommended guidelines.  However, a more realistic measure is that 3 were 1 week late and 2 were 2 weeks late.  This gives us 1 week by 5 days per week by 3 patients plus 2 weeks by 5 days per week by 2 patients = 35 patient-days-late.  Is this bad?  Of course it is, it should be zero.  We can add this measurement to our system.

So unit-days-late is another measure that we can use to evaluate the performance of a subsystem with – anywhere that there is a clear hand-off to another subsystem.

If the subsystems are aligned to the goal of the system we should expect patient-days-wait to decline and patient-days-late to be zero.  Now, these measures are excellent at monitoring subsystems – nodes in the waiting list network for instance – but there is no reason why they can not be used for the whole system as well.  If we use them for the whole system, maybe divided by specialization, then they also provide us with a non-financial measure of system success.  They don’t measure wellness in the community directly but rather indirectly as the absence or decrease in unwellness.  We should strive to reduce the unwellness, wouldn’t you agree?

Let’s hope that one day we can see in district health board meetings a 12-24 month running graph tabled for each major subsystem showing patient-days-wait and patient-days-late.  Then we will know at a glance whether we are all moving in the right direction or not.

We can test for obfuscation with a simple graph.

In the graph below we have some initial criteria for admission to an elective list – patients who have managed to reach the “access threshold.”  Over time the total number of patient-days-wait increases as the effects of system dependency, variability, and an absence of knowledge of how to protect the constraint cause output to be lower than input into the list.

At some time the length of wait and the number of patients waiting becomes too great.  There is a reassessment of the “access threshold” and a limit is imposed.

The limit is imposed via new criteria for the access threshold.  Some of the previous patients are “parked” in new categories such as the “residual waiting list.”  Nevertheless, the patient-days-wait continues to increase as before, and for the same reasons, but now artificially depressed for a time by the adjustment.

Patient-days-wait increases, that is until, once again, the length of wait and the number of patients waiting becomes too great.  There is another reassessment of the “access threshold” and a new limit is imposed.