Sir, we read this interesting paper on the complex subject of NAFLD by Sattar, Forrest and Preiss. Unfortunately their clinical algorithm contains several important errors we would like to elucidate :-
1. The contents of the two Statement boxes at the start of the Algorithm seem arbitrary. What about the inclusion of choices for all the logical combinations of variables such as “symptomatic / asymptomatic”, “fatty liver on US / no fatty liver on US” and “normal / abnormal LFTs”?
2. The right-hand rectangular box displaying the text “if AST:ALT > 0.8, consider specialist referral” contains a decision ( resulting from a question ) followed by a process. These different flow chart components cannot be included in the same box and need separating ( q.v. re-designed algorithm below ). Decisions should be displayed in diamond-shaped boxes and processes in rectangular-shaped boxes according to the BSI ( British Standards Institute ) Standard BS4058 for flow charts [ 1, 2 ].
3. The algorithm will fail if the patient’s AST:ALT ratio is equal to 0.8 because the algorithm has no selectable choice for this value.
4. The algorithm uses the numeric pointers “1”, “2” and “3” and a large amount of text to transfer control from the bottom of the algorithm to the middle of it. A clinical algorithm in the form of a flow chart is ideally suited to displaying cyclical data flow using arrow-headed, data-flow lines without the need for descriptive text or pointers. Pointers used in this context are analogous to the use of the GOTO statement in the computer language, BASIC, which encourages “unstructured” programming.
We understand that designing a complex cyclical algorithm like this one is a difficult job but we feel that our re-designed version shown below clarifies the decision-making logic and data flow using “direct wiring” to replace the repetitive text describing the cyclical data flow within it. The re-designed algorithm also disambiguates the display of the three types of results from the patients’ clinical reviews for each of the three levels of ALT enzyme abnormality into “Better”, “Static” and “Worse” and the different clinical interventions that need to be followed.
References
1. Colman A, Richards B.
Management of Pneumothorax. Clinical algorithms should be standardised.
British Medical Journal, 307, 443, 1993. http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1678411/
( accessed August 2014 )
2. Colman A, Athey R, Richards B.
Flawed logic, duplicate and missing data make the opioid algorithm in Figure 2 unusable.
British Medical Journal [ eLetter ], 02 July 2014. http://www.bmj.com/content/346/bmj.f2174/rr/759689
( accessed August 2014 )
Competing interests:
No competing interests
03 August 2014
Andrew W Colman
Clinical Computer Scientist and Staff Physician
Bernard Richards, Emeritus Professor of Medical Informatics, The University of Manchester, Oxford Road, Manchester, M13 9PL
Gateshead NHS Foundation Trust
Queen Elizabeth Hospital, Gateshead, NE9 6SX - E-mail acolman@clara.net
Rapid Response:
Sir, we read this interesting paper on the complex subject of NAFLD by Sattar, Forrest and Preiss. Unfortunately their clinical algorithm contains several important errors we would like to elucidate :-
1. The contents of the two Statement boxes at the start of the Algorithm seem arbitrary. What about the inclusion of choices for all the logical combinations of variables such as “symptomatic / asymptomatic”, “fatty liver on US / no fatty liver on US” and “normal / abnormal LFTs”?
2. The right-hand rectangular box displaying the text “if AST:ALT > 0.8, consider specialist referral” contains a decision ( resulting from a question ) followed by a process. These different flow chart components cannot be included in the same box and need separating ( q.v. re-designed algorithm below ). Decisions should be displayed in diamond-shaped boxes and processes in rectangular-shaped boxes according to the BSI ( British Standards Institute ) Standard BS4058 for flow charts [ 1, 2 ].
3. The algorithm will fail if the patient’s AST:ALT ratio is equal to 0.8 because the algorithm has no selectable choice for this value.
4. The algorithm uses the numeric pointers “1”, “2” and “3” and a large amount of text to transfer control from the bottom of the algorithm to the middle of it. A clinical algorithm in the form of a flow chart is ideally suited to displaying cyclical data flow using arrow-headed, data-flow lines without the need for descriptive text or pointers. Pointers used in this context are analogous to the use of the GOTO statement in the computer language, BASIC, which encourages “unstructured” programming.
We understand that designing a complex cyclical algorithm like this one is a difficult job but we feel that our re-designed version shown below clarifies the decision-making logic and data flow using “direct wiring” to replace the repetitive text describing the cyclical data flow within it. The re-designed algorithm also disambiguates the display of the three types of results from the patients’ clinical reviews for each of the three levels of ALT enzyme abnormality into “Better”, “Static” and “Worse” and the different clinical interventions that need to be followed.
References
1. Colman A, Richards B.
Management of Pneumothorax. Clinical algorithms should be standardised.
British Medical Journal, 307, 443, 1993.
http://www.ncbi.nlm.nih.gov/pmc/articles/PMC1678411/
( accessed August 2014 )
2. Colman A, Athey R, Richards B.
Flawed logic, duplicate and missing data make the opioid algorithm in Figure 2 unusable.
British Medical Journal [ eLetter ], 02 July 2014.
http://www.bmj.com/content/346/bmj.f2174/rr/759689
( accessed August 2014 )
Competing interests: No competing interests