• Users Online: 148
  • Home
  • Print this page
  • Email this page

   Table of Contents      
Year : 1999  |  Volume : 47  |  Issue : 1  |  Page : 41-48

An introduction to clinical decision analysis in ophthalmology

1 Schell Eye Hospital, Department of Ophthalmology, Christian Medical College, Vellore, India
2 Department of Community Health, Christian Medical College, Vellore, India

Correspondence Address:
Ravi Thomas
Schell Eye Hospital, Christian Medical College, Arni Road, Vellore - 632 001
Login to access the Email id

Source of Support: None, Conflict of Interest: None

PMID: 16130287

Rights and PermissionsRights and Permissions

Ophthalmologists are often confronted with difficult clinical management problems. In such cases, even published experience may be limited; consequently multiple, generally unproven management options are usually available. When placed in such situations, most of us decide on the most appropriate course of action based on intuition or (limited) previous experience. In this article, we use examples to introduce the concept of decision analysis, a method of generating objective decisions for complex clinical problems.

Keywords: Decision analysis, ophthalmology, clinical decision making

How to cite this article:
Korah S, Thomas R, Muliyil J. An introduction to clinical decision analysis in ophthalmology. Indian J Ophthalmol 1999;47:41-8

How to cite this URL:
Korah S, Thomas R, Muliyil J. An introduction to clinical decision analysis in ophthalmology. Indian J Ophthalmol [serial online] 1999 [cited 2023 Jan 29];47:41-8. Available from: https://www.ijo.in/text.asp?1999/47/1/41/22809

Click here to view

Click here to view

Click here to view

Click here to view
There are several methods of making clinical decisions. While satisfactory clinical decisions can be based on (years of) experience, good decisions, especially in complicated cases, may involve a great deal more. This includes knowledge of the evidence in favour of a particular pattern of management, as well as awareness of the competing risks and benefits. Decisions become even more difficult when an available treatment poses substantial risks as the price of benefits. Moreover, in such situations, valid evidence for management options is difficult to come by and the decision may become quite arbitrary.

While most ophthalmologists may feel comfortable making an intuitive decision based on "experience", in a complicated case such experience is usually limited. In such cases, hard figures obtained by using all available evidence may make decision making easier. In recent years, a variety of methods of "quantitative" decision-making have become popular. These include decision analysis, cost-benefit analysis and cost-effectiveness analysis.[1] In these methods, the decision-making process is presented in an explicit manner, assigning numerical values to the components of the problem. The consequences of each option can then be examined quantitatively and the result with the highest value selected.

In this article we use simple ophthalmic examples to introduce the reader to the application of decision analysis in clinical settings. Difficult decisions are systematically subjected to the strategy of decision analysis, which usually helps solve the problem in a more scientific manner. This method has been described for clinical problems in ophthalmology as well as other medical specialties.[2],[3]

  Definition Top

Decision analysis is defined as a method of describing intricate clinical problems in a clear-cut manner, identifying all potential courses of action (both diagnostic and therapeutic), weighing the probability of occurrence of each outcome with each course of action and assigning a value (or utility) for all the possible outcomes.[4] Simple calculations are then made and the optimal course of action selected. In other words, all relevant courses of action and their consequences (based on available evidence) are mapped out and evaluated in order to identify the single course of action that may best serve the patient. While this sounds wonderful, unfortunately, as we all know, in real life there may not actually be a "single best course of action." It is precisely in such difficult situations that decision analysis may come in handy.

  The Decision Tree Top

The basic tool for decision analysis is the decision tree diagram. Here the problem, the various courses of action available (acts), the possible events that can affect each act, and the outcomes of each course of action are conveniently mapped out.

  A real life concept Top

To introduce the concept, let us consider a situation faced by postgraduate trainees in our department. On Wednesdays and Saturdays, the department has a Grand Round, a formal teaching and management round of the wards. All cases are presented by the bed-in-charge (the trainee), who has worked up the patient; all trainees are expected to be aware of all patients in the ward, but know all about the "problem" cases as well as the ones with interesting clinical findings. The rounds can sometimes be quite fiery, posing challenges to all trainees; the probability of depression following such a round is fairly high.

Our trainees, like the average trainee elsewhere, also enjoy extracurricular activities and late-night movies. The decision to go to a late-night movie on the night preceding the grand round has to be weighed against the chances of not knowing the answers to questions pertaining to problem cases in the ward. The consequences may be depression. To attend the latest release the night before the grand rounds or not? That is the question. The acts in this case are two:

  1. 1. Go to the movie and hope that questions are not asked, or are directed at other trainees.

  2. 2. Play it safe; do not go to the movie.

The events which would influence the outcome of each act, would in this case be:

  1. 1. The questions are unfortunately directed at the movie-going trainee.

  2. 2. No questions are directed at him.

We can see that acts are under the control of the trainee, while the events are not. The outcome resulting from each act depends on the event. The choice of acts is shown as a fork with a separate branch for each act.

The events are represented by branches in separate forks. An example of decision tree is shown in [Figure - 1]. As the decision-tree diagram can become quite elaborate, specific symbols are utilized to make it easier to follow. The points where the tree branches out are called nodes. Square nodes are used to represent the action fork, that is, forks of the tree where the clinician can make a choice and has some control. These are appropriately called choice nodes. Round nodes are used to represent the event fork, i.e., one that is not under the clinician's control (chance nodes).

The flow in such a diagram should be chronological from left to right, i.e., the act is followed by the event, which is followed by the outcome. The outcome resulting from a particular act-event combination has to be shown at the right end position of the corresponding path. The analysis need only include every alternative that the decision-maker wants to consider, therefore, only relevant choices are incorporated in the tree. For instance, our trainee might decide to go to the movie and then take leave the next day. However, this alternative (with even more painful consequences) was not considered and so does not enter the decision tree diagram.

  Additional Elements in Decision Making Top

Once we have constructed our decision-tree diagram, we need to add two more elements to it: (1) Payoffs or costs associated with the outcome, and (2) Event probability.

  1. Payoffs Top

These are represented by a number that denotes the worth of the outcome. After ranking, we quantify the payoff values on a simple ranking scale between 0 to 1 as follows:

The decision tree after entering the payoff values is shown in [Figure - 1]. A more conscientious student may prefer to rank the outcomes differently. In a clinical situation, a number is assigned to the outcome after taking into account the means of attaining this outcome, the quality of the outcome and the price if any, to attain this. This number or at least a "guesstimate" of it is usually available from the literature, or it may need to be derived in one of the following ways.

  1. a. The treating clinician may decide on what he/she thinks would be an appropriate rating.

  2. b. A group of clinicians interested in the case may want to sit together and decide a collective value.

  3. c. The concerned patient may be brought into the decision-making process and a rating from the patient's perspective obtained.

  4. We feel the patient should always be involved in the process.

      2. Event Probability Top

    This is the other factor inserted into the decision tree. It denotes the chances or likelihood of a particular event happening. This number is usually available from the literature, such as, Success /Failure /Complication Rates of a particular surgery (as shown in [Figure - 2].

    In our example, if the "problem" patient is on the first bed, the probability of a detailed discussion and therefore of more questions being asked is much higher (say, 90%) than if this patient is on the last bed where everyone is in more of a hurry to finish and start the outpatient clinic (say, 10%). In real life this could be different if the grand round starts at the last bed.

    For the purposes of our example, let us assume that this patient is on the 40th bed (out of 100 patients). We might put the probability of being asked a question at around 40%. The tree after ranking and entering all the probability and payoff values is shown in [Figure - 3]. To come to our decision, we now go through the procedure of "folding back". Here, we multiply all the values along a particular arm from right to left (from utilities backward) till we reach a node. We then add together all the values that reach the node and then continue the same process till we reach the last act node, that is the node that decides the issue. As it is inappropriate to multiply percentages, for the calculations all the percentages are converted to decimal figures.

    In the example given, there is only one node before the deciding act node, and so the calculation would be:

    (0.5 x 0.4) + (1.0 x 0.6) = 0.8 - Go for movie

    (0.75 x 0.4) + (0.0 x 0.6) = 0.3 - Play it safe

    In this case the decision would be to go to the movie.

      A Common Ophthalmological Problem Top

    Now that our readers have (hopefully) understood the concept of decision analysis, let us apply it to a common ophthalmological problem.

    Consider a general cataract-screening program in a village. A "complete" external flash-light examination is done and a Schiotz intraocular pressure (IOP) is measured. However, the medical retina consultants insist that we should use this opportunity to screen for diabetic retinopathy also. The question we want to settle is whether to dilate all the patients in order to identify at least severe diabetic retinopathy that needs laser treatment to prevent blindness. This would mean extra time spent in dilating all eyes and examining the fundus in detail (perhaps even a biomicroscopic fundus examination using a slitlamp taken into the field).

      The six steps Top

    To take you through this decision-making strategy we will use the 6 steps described by Sackett.[5]

    Step 1. Create a decision tree or map of all the pertinent courses of action and their consequences. The tree for our example is as seen in Figure 4.

    Step 2. Assign probabilities to the branches that sprout from each chance node.

    A literature search has shown various studies done world wide that provides the prevalence of diabetes mellitus. [6,7] A study from Vellore (rural population) suggests an age-adjusted prevalence of about 3.2% in males and 1.3% in females.[8] Data obtained from diabetic patients have shown that severe diabetic retinopathy requiring treatment is around 20%.[6] Hence the probability of a person being screened and found to have severe diabetic retinopathy requiring treatment is:

    3/100 x 20/100 = 60/10000, or 0.06%

    Entering this information into the tree makes it look like [Figure - 5].

    Step 3. Assign "utility" values to each of the potential outcomes.

    Obviously some health outcomes in the decision tree are better than others because the qualitative and quantitative benefits of these outcomes are better. This product (quality x quantity) in epidemiological jargon is called the utility of the particular outcome. It may make things easier to think of it as "value" or "worth". Many of us already think in these terms while making most clinical decisions (for example, this patient will probably see better for longer with this particular treatment). Before deciding on the actual values, let us rank the various outcomes in the tree from best to worst.

    For what is primarily a cataract screening camp, we thought that the following ranking might be appropriate:

    Best 1. Saved time and money, i.e., the patient had no retinopathy and was not dilated. The happy patient did not give an adverse report of the camp and hence did not prevent other cataract blind from attending the next day.

    Best 2. Prevention of blindness, i.e., the patient had severe retinopathy and was detected after dilatation.

    Best 3. "Waste" of time and money, i.e., normal patient was dilated.

    Worst 4. Blindness, i.e., severe retinopathy missed because patient was not dilated.

    After ranking, we can now proceed to assign numerical values to the outcomes. This is the most difficult part because we are now forced to be explicit in assigning these utilities. We may be able to get some help from the literature (morbidity/mortality rates). Also we may ask the patient to rank and assign these values. Obviously the patient must be made aware of the potential advantages and consequences of each action.

    For our example, we felt the following utility values were appropriate.

    We now add these values to our tree and it looks like [Figure - 6].

    Step 4. The decision tree is now "folded back" by combining the probabilities and the utilities for each node from right to left, multiplying the values along each limb and then adding up all the values at each node. Remember, we need to convert all our values to decimal figures for the calculation.

    1. 0.0006x0.75 = 0.00045

    2. 0.9994x0.50 = 0.4998

    0.00045 + 0.4998 = 0.50025 (value for decision to dilate).

    1. 0.0006x0.00 = 0

    2. 0.9994x1.00 = 0.9998

    0.00 + 0.9998 = 0.9998 (value for decision not to dilate).

    Step 5. Choose the decision with the highest score.

    The decision is not to dilate all patients seen in the camp.

    Step 6. Sensitivity Analysis. This is, however, not the end of our decision-making process, and the sixth step is perhaps the most crucial. This step is the sensitivity analysis. Our decision must be tested for its vulnerability to clinically sensible changes in probabilities and utilities.

    For example, what if the prevalence of diabetes was 5% and that of severe retinopathy 40%? We used these values of 5% (instead of 3%) for the prevalence of diabetes mellitus and 40% (instead of 20%) for severe diabetic retinopathy and still arrived at the same decision: patients in our screening camps are not dilated. As a bonus, the decision tree helped convince our medical retina consultants.

    Decision analysis is not complete without sensitivity analysis. If the decision arrived at remains the same even after credible changes in probabilities and utilities are done, then the decision is most probably the best option available to the patient. If however it changes, then we are in a situation where it probably makes little difference whichever decision we choose. In fact it has been found that this kind of "toss up" situation occurs in about one-sixth of all clinical decision analyses.[9] This is actually quite good, considering that this method is generally used for difficult decisions and applying this technique does provide definite answers in the remaining five-sixths of the cases.

      Decision Making in Problem Cases Top

    The above examples illustrate the concept of decision analysis, but are not really difficult enough to warrant it. The decisions could have been made on pure common sense alone. Decision analysis is a method of using the available facts to quantify our common sense. Let us now consider a more controversial problem; the type where decision analysis might really make a difference.

    Mr. K is a 68-year-old man who sustained injury to his left eye with his walking stick 15 days previously. He complained of decreased vision in his left eye. Although there had been some pain and redness for 3-4 days post trauma, he now had no pain.

    On examination the best-corrected vision in his uninjured right eye was 6/60. The vision in his left, injured eye was 3/60, improving to 6/12 with the appropriate aphakic spectacle correction. Examination of the right eye was significant only for a grade 4 nuclear sclerosis with posterior subcapsular cataract. The left eye was aphakic. A fundus examination revealed a dislocated intact lens in the inferior periphery with a few intact zonular attachments at its 6 o'clock equator. The retina was otherwise normal. The IOP was 15 mmHg in the right eye and 14 mmHg in the left. Gonioscopy revealed open angles in both eyes; there was no evidence of angle recession in the left eye.

    The appropriate management of such a case is controversial. It is generally accepted that Mr. Edmund Hillary's explanation for why he climbed Mount Everest: "because it is there", is probably not enough justification for removal of a dislocated crystalline lens. We do not generally remove a dislocated crystalline lens in the vitreous just "because it is there". Therefore we had a decision to make. Should we proceed with an expensive operation to remove the dislocated lens? On the other hand, he was asymptomatic with the distinct possibility of not suffering any ill effects for the remaining duration of his life (the average life expectancy of a male in Tamil Nadu is 61 years). Should we then provide aphakic glasses or contact lenses instead and follow him up?

    The decision tree that we constructed is shown in [Figure - 7].

    While searching the literature for experience in such cases, we defined "success" as a final vision of 6/60 or better. The tree became bigger when we considered that once we committed ourselves to surgery we are obliged to go all out for the eventual success of the management. We also added one more option: utilize available finances to render him pseudophakic in his uninjured right eye with state-of-the-art cataract surgery.

    For the purposes of this introduction to decision analysis, we limited our tree to that already shown.

    To keep things simple, we wanted to use the utility values of "1" for "success" and "0" for "failure". However, on talking to the patient and presenting the various courses of action available to him, he decided on the following ranking and utility values based on his financial status and convenience.

    Best: 1. No treatment-No complications-Success 1.00

    2. Surgery- No complications- Success - 0.75

    3. No treatment-Complications-Treatment-Success - 0.60

    4. Surgery-Complications-Treatment-Success -0.40

    5. No treatment-Complications-Treatment-Failure - 0.20

    Worst: 6. Surgery-Complications-Treatment-Failure- 0.00

    The probability values for the above scenarios were obtained from the literature and also entered into the tree.[10],[11]

    Chances of complications if left alone - 25-30%

    Chances of complications if primary lens removal is done -15%

    The complications in both cases refer to retinal detachment and secondary glaucoma.

    Chances of success of surgery done to correct a complication occurring if left alone - 75%

    Chances of success of surgery done to correct a complication following primary lens removal - 60%

    The completed tree after entering all the probability and payoff values is shown in [Figure :8].

      Folding back the tree: Top

    As is seen, the decision tree in this case suggested we should leave the traumatized eye alone, at least for the present.

    But wait. We have almost forgotten one important part of the strategy: to ensure that our decision withstood credible changes in the assumed probability values, we still needed to perform a sensitivity analysis. We re-did our calculations using the following changes in the probability values; as you will realize, all the values are now weighted towards primary lens removal.

    Chances of complications if left alone - 45%

    Chances of complications if primary lens removal done - 5%

    Chances of success of surgery done to correct a complication occurring if left alone - 60%

    Chances of success of surgery done to correct a complication following primary lens removal - 80%

    This is in favour (albeit less forcefully) of our decision to leave well enough alone. The patient was happy with our decision and went home with his aphakic glasses. He did not want the cataract in his right eye removed just yet. One year later his condition was still status quo.

    We hope that we have been able to demonstrate the usefulness of decision analysis. In a complicated situation where several lines of "correct" management exist and where different, experienced ophthalmologists may offer strongly held but different, opinions we find it a convenient and practical method of making a decision. The value of the competing options can be quantified and assessed scientifically. In addition, identifying toss-up situations will help decrease the many heated, but often inconsequential debates that tend to occur in the course of making the decision best suited to the individual patient.

      References Top

    Fletcher RH, Fletcher SW, Wagner EH. Clinical Epidemiology. The Essentials. 2nd ed. Baltimore:Williams and Wilkins; 1988. p 69-73.  Back to cited text no. 1
    Kassof J, Meyer DR. Early office-based vs late hospital-based nasolacrimal duct probing: A clinical decision analysis. Arch Ophthalmol 1995;113:1168-71.  Back to cited text no. 2
    Mair J, Smidt J, Lecchleitner P, Diens HF, Puschendorf B. A decision tree for the early diagnosis of acute myocardial infarction in non-traumatic chest pain patients at hospital admission. Chest 1995;108:1502-9.  Back to cited text no. 3
    Sackett DL, Haynes RB, Guyatt GH, Tugwell P. Clinical Epidemiology: A Basic Science for Clinical Medicine. 2nd ed. Toronto:Little Brown and Co; 1991. p 238.  Back to cited text no. 4
    Sackett DL, Haynes RB, Guyatt GH, Tugwell P. Clinical Epidemiology: A Basic Science for Clinical Medicine. 2nd ed. Toronto: Little, Brown and Co; 1991. p 140-45  Back to cited text no. 5
    Baker RS, Watkins NL, Wilson MR, Bazargan M, Flowers CW Jr. Demographic and clinical characteristics of patients with diabetes presenting to an urban public hospital ophthalmology clinic. Ophthalmol 1998;105:1373-79.  Back to cited text no. 6
    Sanchez, Thorin JC. The epidemiology of diabetes and diabetic retinopathy. Int Ophthalmol Clin 1998;38:11-18.  Back to cited text no. 7
    ICMR Collaborative study on Coronary Heart Disease, Vellore Centre, Monograph 1994, Vellore, Christian Medical College, Tamil Nadu, India.  Back to cited text no. 8
    Kassiver JP, Pauker SG. The toss-up. N Engl J Med 1981;305:467.  Back to cited text no. 9
    Jarret WH. Dislocation of the lens; A study of 166 hospitalized cases. Arch Ophthalmol 1967;78:289-96.  Back to cited text no. 10
    Lewis H, Blumenkranz MS, Chang S. The treatment of dislocated crystalline lens and retinal detachment with perfluorocarbon liquids. Retina 1992;12:299-304.  Back to cited text no. 11


      [Figure - 1], [Figure - 2], [Figure - 3], [Figure - 4], [Figure - 5], [Figure - 6], [Figure - 7], [Figure - 8], [Figure - 9], [Figure - 10]

      [Table - 1], [Table - 2]


    Similar in PUBMED
   Search Pubmed for
   Search in Google Scholar for
 Related articles
    Access Statistics
    Email Alert *
    Add to My List *
* Registration required (free)  

  Additional Eleme...A Common Ophthal...Decision Making ...Folding back the...
  In this article
The Decision Tree
A real life concept
1. Payoffs
2. Event Probability
The six steps
Article Figures
Article Tables

 Article Access Statistics
    PDF Downloaded4    
    Comments [Add]    

Recommend this journal