2. Examples

This chapter presents the inner workings of ABEL with the help of annotated examples. In this chapter the reader is not expected to understand how the program accomplishes its task, but rather just what it does. The succeeding chapters will examine the structure of the program and the method by which each step is accomplished. We will consider two examples: (1) a patient suffering from moderately severe salmonellosis, and (2) a patient suffering from moderately severe salmonellosis and vomiting. The selection of the medical examples is motivated by our desire to make the medical contents of the examples as simple as possible. In chapter 6 we will revisit these examples and discuss how the program accomplishes each of its tasks.

2.1 Example 1: Salmonellosis

For the first example let us consider a 40 year old 70 Kg male patient who has been suffering from moderately severe salmonellosis and, as a result, has developed moderately severe metabolic acidosis and hypokalemia. To illustrate the program let us provide it initially with only the laboratory analysis of the patient's blood sample (serum analysis) without any clinical information.

Based on these data, the program generates all possible acid-base disturbances that can account for the laboratory data. It then prunes and rank-orders these disturbances based on their complexity, likelihood and severity of each component. The rank-ordered list of likely disturbances is:

---- Patient Acid-Base Profile ----
1. metabolic-acidosis [severity: 0.4] very likely 
2. chronic-respiratory—alkalosis [severity: 0.68] 
   + acute-respiratory—acidosis [severity: 0.32] unlikely

Fig. 4. Graphic depiction of the two Acid-Base hypotheses

The computation of the acid-base profile is based on the Nomogram of Acid-Base Disturbances described in chapter 1. Figure 4 shows the relevant region of this nomogram with the loci of the two hypothesized disturbances. The estimation of the severity of a disturbance is based on the length of the segment along the locus of that disturbance. Thus, we note that the severity of the single acid-base disturbance (metabolic acidosis) is only about 0.4 while an equivalent acid-base disturbance composed of chronic respiratory alkalosis and acute respiratory acidosis has seventies of 0.68 and 0.32 respectively.⁷

Next, the program creates a PSM for each possible acid-base disturbance and interprets the laboratory data in the context defined by each acid-base disturbance. For example, with the assumption of fully compensated metabolic acidosis, the entire change in the PCO2 may be considered chronic, therefore, the chronic component of the PCO2 will be 30 meq/l, while with the assumption of chronic respiratory alkalosis and acute respiratory acidosis, the chronic component of PCO2 is due only to the chronic component of this disturbance, therefore reading from the nomogram we find that the chronic value of PCO2 in this case will be approximately 16 meq/l.

The program then aggregates its patient-specific physiologic knowledge to formulate an interpretation of the laboratory data at the clinical level. The computer generated explanation of its interpretation of these data under the two major hypotheses is described in figures 5 and 6.

A quick look at the two clinical level explanations shows that the structure involving hypokalemia and acidemia is common to the two hypotheses. They differ in their accounting for acidemia. Note that the clinical level abstraction of the two hypotheses is fairly simple in structure and does not contain any feedback cycles. The cycles present at the intermediate level describing the interaction between the acidemia, hypobicarbonatemia and hypocapnia have been abstracted away. A closer look at these feedback cycles shows the principal difference between the two hypotheses. In the first case, the change in the acid-base state is a consequence of loss of HCQ3 from the body which causes hypobicarbonatemia, whereas in the second it enters as primary disturbance in ventilation which alters the PCO2. Finally, we note that the first hypothesis has two unaccounted findings while the second hypothesis contains three unaccounted findings.

In the context of this initial analysis of the patient's condition, the program starts the diagnostic exploration. An annotated (in italics) transcript of the program's diagnostic behavior is shown next.

The program computes the diagnostic closures for the two hypotheses and decides to pursue the first hypothesis.

Differentiating between the causes of the leading
complete hypothesis.
1 SALMONELLOSIS
2 URETEROSIGMOIDOSTOMY
3 VILLOUS-ADENONA
------------------------
4 DISTAL-RTA
5 PROXIMAL-RTA
6 ACUTE-RENAL-FAILURE
7 CHRONIC-RENAL-FAILURE
continue? ==>

The list above contains all possible diseases that can explain some part of the first hypothesis. The list is divided into groups of diseases by the number of unaccounted findings that each disease can explain succinctly. Within each group the diseases are ordered by a secondary scoring criterion based on the quality of their match with the hypothesis and their potential to be ultimately confirmed.

Differentiating between
SALMONELLOSIS URETEROSIGMOIDOSTOMY VILLOUS-ADENOMA

Does the patient have any of the following? 
1 SALMONELLOSIS 
2 URETEROSIGMOIDOSTOMY 
3 VILLOUS-ADENOMA
Please enter numbers between 1 and 3
Present: ==> E(xplain) 

The user indicates that he would like an explanation. The program prompts by indicating in a menu the possible choices for explanation.

1 DESCRIBE-HYPOTHESIS
2 DESCRIBE-DIAGNOSTIC-REASONING
3 DESCRIBE-FINDING
4 DESCRIBE-CAUSE
5 DONE
==> 2
The stack of diagnostic goals:

3 DIFF-CAUSES
2 CONFIRM-CH
1 DIAGNOSE

I am diagnosing the possible acid base disturbances. I am
confirming the leading hypothesis. I am differentiating between
its leading causes.
1 DESCRIBE-HYPOTHESIS
2 DESCRIBE-DIAGNOSTIC-REASONING
3 DESCRIBE-FINDING
4 DESCRIBE-CAUSE
5 DONE
==> 4

Please enter a cause for viewing: ==> salmonellosis

Salmonellosis may account for moderate metabolic acidosis and
moderate hypokalemia as follows.
Moderate salmonellosis may cause metabolic acidosis.
Moderate salmonellosis may cause hypokalemia
1 DESCRIBE-HYPOTHESIS
2 DESCRIBE-DIAGNOSTIC-REASONING
3 DESCRIBE-FINDING
4 DESCRIBE-CAUSE
5 DONE /
==> 5

Does the patient have any of the following?
1 SALMONELLOSIS
2 URETEROSIGMOIDOSTOMY
3 VILLOUS-ADENOMA
Present: ~> none Absent: ==> none Unknown: 1 2 3

Differentiating between

SALMONELLOSIS URETEROSIGMOIDOSTOMY VILLOUS-ADENOMA

Does the patient have one of the following?
1 DEHYDRATION
2 EDEMA
Present: ==> none Absent: ==> none Unknown: 1 2

The program is expecting dehydration. Therefore, when we fail to confirm or deny the dehydration the program pursues the finding further.

I would like to ask about the effects of SALMONELLOSIS.
Is the value of SERUM-CREATININE known? ==> E(xplain)

High serum creatinine may be caused by moderate salmonellosis
as follows:

Moderately high serum creatinine may be caused by moderate
dehydration, which may be caused by salmonellosis.

The user indicates that he would like a justification for this question. The program generates the explanation by tracing back the causal path from serum creatinine in the DC associated with the goal of the question.

The program's diagnostic reasoning at this point can be described as follows: The top level goal of the program is to do diagnosis. In order to do the diagnosis the program evaluates the two alternate hypotheses and selects the first hypothesis (metabolic-acidosis) for confirmation. To confirm this hypothesis it selects the set of leading causes for the first hypothesis, namely, salmonellosis, ureterosigmoidostomy and villous-adenoma. It then attempts to differentiate between these causes. It determines the findings predicted by each of these three causes, orders them according to their discriminatory power, and asks about them, the most discriminating finding first.

Is the value of SERUM-CREATININE known? ==> yes

Please enter the attributes of SERUM-CREATININE
What is the VALUE of SERUM-CREATININE ? ==> 3
What is the START-TIME of SERUM-CREATININE ? ==> 0

Is the value of MEAN—ARTERIAL-BLOOD-PRESSURE known? ==> yes

Please enter the attributes of MEAN-ARTERIAL-BLOOD-PRESSURE

What is the VALUE of MEAN-ARTERIAL-BLOOD-PRESSURE ? ==> 75

The program has now completed one full cycle of its planned diagnostic inquiry. It now incorporates this information into both hypotheses and starts the next cycle of diagnostic planning.

Starting next cycle of diagnosis

Differentiating between the causes of the leading complete hypothesis.
1 SALMONELLOSIS
2 VILLOUS-ADENOMA
3 URETEROSIGMOIDOSTOMY
4 ADRENAL-INSUFFICIENCY
5 DIABETES-INSIPIDUS
6 ACUTE-RENAL-FAILURE
7 CHRONIC-RENAL-FAILURE

8 DISTAL-RTA
9 PROXIMAL-RTA
continue? ==>

The program has already gathered sufficient information to confirm salmonellosis. It is unable to do so because we have not implemented the criteria for confirming a disease yet. However, we note that the information gathered has provided substantial categorical separation between the gastrointestinal and the renal causes of the metabolic-acidosis with hypokalemia.

Pursuing SALMONELLOSIS: no findings available

Differentiating between VILLOUS-ADENOMA URETEROSIGMOIDOSTOMY:
    no finding available

Differentiating between ADRENAL-INSUFFICIENCY DIABETES-INSIPIDUS
ACUTE-RENAL-FAILURE CHRONIC-RENAL-FAILURE

As there are more than two items in the differentiation set, the program groups these items into renal and extra-renal sets. It then pursues the renal set first.

Does the patient have any of the following?
1 CHRONIC-RENAL-FAILURE
2 ACUTE-RENAL-FAILURE
Present: ==> none    Absent: ==> 1 2

Does the patient have any of the following?
1 ADRENAL-INSUFFICIENCY
2 DIABETES-INSIPIDUS
Present: ==> none    Absent: ==> 1    Unknown: 2

Differentiating between DISTAL-RTA PROXIMAL-RTA

Does the patient have any of the following?
1 PROXIMAL-RTA
2 DISTAL-RTA
Present: ==> none    Absent: ==> none    Unknown: 1 2

The program has now completed the second cycle of its planned diagnostic inquiry. No new finding was uncovered during this cycle of diagnosis and the retore, the program does not need to revise The two hypotheses. However, during the planning of the next cycle of diagnostic inquiry it discovers that all the questions relevant to the etiology of metabolic-acidosis and hypokalemia have already been exhausted. Therefore, in order to proceed it requests that some possible etiology be assumed.

Starting next cycle of diagnosis
No new finding added in the previous cycle.

All possible etiologies that could explain the patient's
illness are unknown. In order to proceed we must at least
hypothetically assume one of them. Possible etiologies that could
explain the patient's illness listed in decreasing order are:
1 SALMONELLOSIS
----------------------------
2 VILLOUS-ADENOMA
----------------------------
3 URETEROSIGMOIDOSTOMY
----------------------------
----------------------------
4 DIABETES-INSIPIDUS
----------------------------
5 DISTAL-RTA
6 PROXIMAL-RTA
Would you like to assume SALMONELLOSIS? ==> yes

Assuming MODERATE ACUTE SALMONELLOSIS.

The program adds salmonellosis to the patient models and re-evaluates two hypotheses before resuming the diagnosis again. However, it immediately recognizes that it has found the missing piece in the puzzle! Based on the assumption that the patient has salmonellosis, the program selects the first PSM (metabolic acidosis) as the explanation of the patient's acid-base disorder. The program's explanations of the successful and the alternate PSMs are shown next. Note, however, that these explanations are provided with the assumption that the patient has salmonellosis, an assumption that needs to be verified.

Diagnosis completed. The successful diagnosis is:

This is a 40 year old 70.0 kg male patient with moderate
salmonellosis. His electrolytes are:

    Na: 142.0     HCO3: 15.0     Anion Gap: 13.0
    K:    3.0     pCO2: 30.0
    Cl: 113.0       pH: 7.32     Creatinine: 3.0

The salmonellosis causes moderate metabolic acidosis and
moderate dehydration. The dehydration causes moderate
hypotension and moderate high creatinine disturbance. The
metabolic acidosis causes mild acidemia. The salmoriellosis and
acidemia cause mild hypokalemia. All findings have been
accounted for.

As can be seen from the above explanation, the program has concluded that salmonellosis provides an adequate explanation for the patient's illness and that the first hypothesis is substantially superior to the alternate hypothesis. The alternate hypothesis with all the available information added is shown below.

This is a 40 year old 70.0 kg male patient with salmonellosis.
His electrolytes are:

    Na: 142.0     HCO3: 15.0     Anion Gap: 13.0
    K:    3.0     pCO2: 30.0
    Cl: 113.0       pH: 7.32     Creatinine: 3.0

The salmonellosis causes moderate metabolic acidosis and
moderate dehydration. The dehydration causes moderate
hypotension and moderate high creatinine disturbance.
Moderate acute respiratory acidosis, moderate chronic
respiratory alkalosis and metabolic acidosis partly
compensate the suspected mild alkalemia leading to the
observed mild acidemia. The salmonellosis and acideniia cause
mild hypokalemia. The chronic respiratory alkalosis and
acute respiratory acidosis remain tO be accounted for. The
alkalemia has only been partially accounted for.

Notice the difference in the two explanations. The first explanation contains only one acid-base disturbance, while the second explanation contains a total of three acid-base disturbances. Furthermore, all the tindlin~s in the first hypothesis hiVE) l)eef I accounted for while the second hypothesis has three acid base distu,bances still to be accounted for.

2.2 Example 2: Vomiting and Salmonellosis

The next example illustrates the program's capabilities in dealing with multiple etiologies, namely salmonellosis and vomiting, which offset the effects of each other on the acid-base balance. We will focus on the program's understanding and its ability to reformulate this understanding. when new information is provided.

Let us consider a patient who is suffering from moderately severe vomiting for the past two days who then develops salmonellosis. Note that the electrolyte and acid-base disturbances in vomiting result from the excessive loss of upper gastrointestinal fluid, while in salmonellosis they result from the loss of lower gastrointestinal fluid. The upper GI fluid is acidic while the lower GI fluid is alkaline, therefore the two tend to have offsetting effects on the acid-base balance. However, vomiting and salmonellosis both cause hypokalemia and dehydration, therefore they compound these effects of each other. For this example, let us consider a patient in which the presentation of vomiting and salmonellosis are such that each exactly cancels the acid-base effect of the other, leaving the patient with no acid-base disturbance. We will illustrate the program's handling of this case by describing the program's understanding of the case at three points during the diagnostic process: (1) just after the electrolyte values are entered in the program, (2) after the finding of vomiting has been presented, and (3) at the end of the diagnostic process.

The program's evaluation of the serum electrolytes and the English explanation of its initial hypothesis are:

Serum Analysis:
    Time: 0
     Sex: male
      Na: 141 meq/l      normal
       K: 2 meq/l        low
      Cl: 108 meq/l      normal
    HCO3: 25 meq/l       normal
    pCO2: 39 mmHg        normal

---- Patient Acid-Base Profile ----
1. normal-acid-base-state

This is a 40 year old 70.0 kg male patient with moderate
hypokalemia. His electrolytes are: ...

The serum analysis reveals only one abnormal finding, hypokalemia. The program starts the diagnostic process by attempting to differentiate between the possible causes of hypokalemia which include vomiting and salmonellosis along with other etiologies such as laxative abuse, diuretic use, hyperaldosteronism etc. The summary of the program's hypothesis after the finding of moderately severe vomiting has been presented is:

This is a 40 year old 70.0 kg male patient with moderate
vomiting. His electrolytes are:
    Na: 143.0    HC03: 25.0    Anion Gap: 12.0
     K: 2.0      pCO2: 39.0
    Cl: 108.0      pH: 7.42

The vomiting causes moderate metabolic alkalosis. Moderate
hypokalemia is partly caused by vomiting leaving some
additional factor causing hypokalemia still unaccounted for. The
hypokalemia and moderate acidemia have only been partially
accounted for.

Notice that the vomiting partially accounts for the observed hypokalemia. However, in order to account for the hypokalemia the program must assume that there has been substantial upper GI fluid loss sufficient to also cause metabolic alkalosis. As this metabolic alkalosis is not consistent with the normal acid-base state the program must decompose the normal acid-base state into offsetting alkalemia and acidemia. The alkalemia which is accounted for by metabolic alkalosis, and acidemia which remains unaccounted for. The remaining unaccounted components now present a picture similar to that of the previous case (example 1) and the diagnosis proceeds similarly. The diagnosis is completed when the program is told about salmonellosis (the remaining disturbance). A summary of the programs' final diagnosis is described next.

This is a 40 year old 70.0 kg male patient with moderate
vomiting and moderate salmonellosis. His electrolytes are:
    Na: 143.0    HC03: 25.0    Anion Gap: 12.0
     K: 2.0      pCO2: 39.0
    Cl: 108.0      pH: 7.42   Creatinine: 3.0

The vomiting causes moderate metabolic alkalosis. The
salmonellosis and vomiting cause moderate dehydration, which
causes moderate hypotension. The dehydration also causes
moderate high creatinine disturbance. The salmonellosis causes
moderate metabolic acidosis. The metabolic acidosis and
metabolic alkalosis cause normal ph. The salmonellosis, normal
ph and vomiting cause moderate acute hypokalemia. All findings
have been accounted for.

The primary focus of this thesis is in developing a methodology for knowledge representation and manipulation that allows our program to exhibit the understanding of patient illness demonstrated above. In the next three chapters we will study in detail this methodology and its implementation before revisiting the same examples again in greater detail.

... on to Chapter 3

This section is part of

Patil, Ramesh S. Causal Representation of Patient Illness for Electrolyte and Acid-Base Diagnosis. MIT Lab for Computer Science Technical Report TR-267. October 1981. Also: Ph.D. Thesis, MIT Dept. of Electrical Engineering and Computer Science.