Transforming Electronic Medical Books to Diagnostic Decision Support Systems Using Relational Database Management Systems

Abstract

The aim of this study was to analyse possibility of developing diagnostic decision support systems (DDSS) from electronic books. Analysis of typical text in major textbooks was performed, and basic structural forms were determined. Knowledge-based DDSS are proposed trying to adapt to the structure of the data and ‘natural’ logic of medical books. We propose a method for transforming a major textbook (with numerous editions) into a simple knowledge-based model of DDSS that authors and editors of the book can easily understand. Structure of the knowledge base includes at least ‘Disease’, ‘Finding’, ‘Relation’ and ‘Thesaurus’ tables. Probability of findings and relations could be expressed as sensitivity, specificity and predictive value. The authors of the book may be kindly asked to use already prepared application and present facts in form of ‘diseases’, ‘findings’ and ‘relations on sets of findings’ when they prepare the text of a book. Maintenance and update of DDSS would be related to the new edition of the book. DDSS would be linked to the electronic text of the book and could be also linked to a local information system. In this way we believe that the proposed model of DDSS would be a stimulus for further development of clinical decision support systems that may be interesting for many readers of the books in which they trust for large number of editions.

Appendix

3.1.1 Example 1

Text from the book

Establishing the Diagnosis of Chronic Kidney Disease (CKD)

The most important initial diagnostic step in the evaluation of a patient is elevated serum creatinine that can help to distinguish newly diagnosed CKD from acute or subacute renal failure. “Previous measurements of plasma creatinine concentration are particularly helpful…” Normal values from recent months suggest that the current renal dysfunction could be more acute, and hence reversible. “In contrast, elevated plasma creatinine concentration in the past suggests that the renal disease represents the progression of a chronic process. Even if there is evidence of chronicity, there is the possibility of a superimposed acute process…” If the history suggests multiple systemic manifestations of recent onset (e.g., fever, polyarthritis, and rash) renal insufficiency can be acute process.

Some of the laboratory tests and imaging studies can be helpful. “Evidence of metabolic bone disease with hyperphosphatemia, hypocalcemia, and elevated PTH and bone alkaline phosphatase levels suggests chronicity. Normochromic, normocytic anemia suggests that the process has been ongoing for some time. The finding of bilaterally reduced kidney size (<8.5 cm in all but the smallest adults) favors CKD.” However, once the CKD is advanced the kidneys are small and scarred.

The above text is extracted from Chap. 280 Chronic Kidney Disease (15, page 2318) that deals with establishing diagnosis. However, some data from previous parts of this chapter may be important for diagnosis (e.g. imaging studies). As we can see there are no probabilities in this text. Because of that some expert has to add it, and the best solution is that the author transforms this text to knowledge base. There is only one disease in this text: chronic kidney disease that is a part of disorders of the kidney and urinary tract. However, metabolic bone disease (part of disorders of bone and mineral metabolism) and normochromic and normocytic anaemia (part of haematology) can be defined as syndromes in this text. There are many findings such as serum creatinine, imaging studies, small and scarred kidneys, hyperphosphatemia, hypocalcaemia, PTH, bone alkaline phosphatase and kidney size.

Now we can acquire previous data in the knowledge base. Diseases and syndromes are acquired in table ‘Disease’ (Table 3.5). Metabolic bone diseases and normochromic and normocytic anaemia are diseases, but here in chronic kidney disease they appear as syndromes. Findings are showed in Table 3.6. Findings are entered without prefix (hyper or hypo) but properties contains these meaning (elevated or decreased). Kidney size are used two times with properties small and normal and enlarged. It is also possible to use small kidney size and normal or enlarged kidney size as different findings.

Table 3.5 Table ‘Disease’ for the first example

Table 3.6 Table ‘Findings’ for the first example

This differentiation is necessary because kidney size has different meaning for the diagnosis of chronic kidney failure and as we can see sensitivity, specificity and predictive values are different. Solving this dilemma in style is on editor. Different meanings have to be explained in Thesaurus table (Table 3.7) where hyperphosphatemia is explained as elevated S-phosphate.

Table 3.7 Table ‘Thesaurus’ for the first example

As we can see from Findings table, diagnosis of chronic kidney disease could be established certainly in two cases: if the kidneys are small and scarred or if S-creatinine is elevated more than 6 months. However, is there some combination of findings that can establish diagnosis certainly? Yes, for example, if we found small kidneys and elevated creatinine for less than 6 months in a person who is not small, then the diagnosis of chronic kidney disease is certain (Table 3.8). Another combination that can raise importance is elevated serum creatinine less than 6 months together with metabolic bone disease and normochromic and normocytic anaemia. If DDSS is connected to information system, sensitivity, specificity and predictive value can be measured.

Table 3.8 Table ‘Relations’ for the first example

Consider now retrieval of knowledge base. If somebody wants to know about findings for the chronic kidney disease, then that is very simple. He will find chronic kidney disease in disease of kidney and urinary tract. Popup with findings will list findings in chronic kidney disease according to sensitivity, specificity and predictive value so it will be similar to table Findings and will include fields finding + property together with syndromes in chronic kidney disease. If query is raised typing the word creatinine, it will list all creatinine findings in all diseases together with properties. If somebody selects elevated creatinine for more than 6 months, diagnosis of chronic kidney disease will appear. If somebody selects elevated creatinine less than 6 months, then findings and properties for chronic kidney disease will appear. However, if somebody selects elevated phosphorus, then some endocrine disease findings will appear in popup. Finding for chronic kidney disease will be at the end of popup. Typing another finding that is specific for kidney diseases will now show another list of findings where kidney disease findings appear first.

3.1.2 Example 2

Text from the book

Confirming the Diagnosis of Acute Rheumatic Fever (ARF)

“Because there is no definitive test, the diagnosis of ARF relies on the presence of a combination of typical clinical features together with evidence of the precipitating group A streptococcal infection…” Experts of the World Health Organization clarified the use of the Jones criteria in ARF. These criteria include a preceding streptococcal type A infection as well as some combination of major and minor manifestations.

The above text is extracted from Chap. 322 Acute Rheumatic Fever (15, page 2755) that deals with establishing diagnosis. However, some data from previous parts of this chapter may be important for diagnosis. As we can see this is special case were diagnostic criteria are clear and could be easily transformed without expert.

There is only one disease – acute rheumatic fever – but several other diseases are used as syndromes: rheumatic chorea, rheumatic carditis, rheumatic heart disease, carditis, polyarthritis, chorea, erythema marginatum, subcutaneous nodules and scarlet fever. There are next findings: fever, polyarthralgia, erythrocyte sedimentation rate (ESR), leukocyte, P-R interval, anti-streptolysin O (ASTO) and throat culture.

Now we can acquire previous data in the knowledge base. Diseases and syndromes are acquired in table Disease (Table 3.9) and findings in table Findings (Table 3.10).

Table 3.9 Table ‘Diseases’ for the second example

Table 3.10 Table ‘Findings’ for the second example

Synonymous are acquired in Thesaurus table (Table 3.11)

Table 3.11 Table ‘Thesaurus’ for the second example

Complex relationship among the findings is showed in table Relations (Table 3.12). Here syndromes are acquired and then relations among findings. There are two levels of relations: primary and secondary that is formed from primary. Because of that, these secondary relations are ‘relations of the relations’ and they have the same format of IDrel in field relations. For example, 10011 is complex and is derived from relations 10008 (necessary one findings-syndrome), 10009 (necessary two findings) and 10010 (necessary one findings). Complex relation 10013 is derived from 10012 which is the same as 10008 but necessary are two findings-syndromes and 10010 (necessary one findings) as can be seen from the text that was transformed.

Table 3.12 Table ‘Relations’ for the second example

Table 215 Criteria for diagnosis ARF (simplified)

Retrieval from this knowledge base is the same as it was discussed in previous example. However, in this case there are more syndromes than in the first example and Relation table is complex. In addition, there are no separate knowledge base tables for some disease. They are unique for all disease as can be seen from ID numbers in the tables but they are showed separately in order to simplify presentation.