Text Generation in MedView Applications

Guide written by Fredrik Lindahl, last updated November 22, 2003

Introduction

This guide describes what rules and definitions exist in the MedView applications for text generation. In the following, the character '\n' defines the new-line character.

Definitions

• Line enders are characters that the generator engine recognizes as enders of a line. Some examples of line enders in the current engine are: . ? ! \n. Note that the newline character is defined as a line ender.

• Colon strings currently consist of the following characters: : -.

• Separators currently consist of only the comma character ','.

• A line starts at the first character that is not whitespace or a line ender character, and ends at the first occurence of a line-ender character. For instance, the line "A camel can walk in the desert." in the following piece of text "\n\nA camel can walk in the desert. The sun shines on the cactus." starts with the 'A' and ends at the dot after "desert".

• A title line is a line that begins on a new line and ends with a new line (and not a line ender, such as '.' or '!'). The first line in the document is considered to be started with a new line. Some examples of title lines: "\nAnamnes\n", "\n\n\nSensitivity\n\n", "Some title\n", where the last example starts at document offset 0 (i.e. the start of the document). Some examples of non title lines: "\nAnamnes.\n", "a previous line. The line to consider.". Title lines have special properties, and are connected with the content following them until the next title line occurs. If all connected content is removed, the title line will be removed as well.

• A significant line is a line that contains significant, non-derived content based on information in a medical record. When a section contains at least one significant line, it will be included in the generated output. For instance, if the following line exists in the template: "The patient bleeds: bleed.", where bleed is a term, and the selected medical record contains content for the bleed-term that does not translate to noline, the line is significant.

• An invalid line is a line that must be removed from output at generation time. An invalid line is produced when a line contains at least one term that does not translate to anything, or if it translated to noline. For instance, if the following line exists in the template: "The patient bleeds: bleed.", where bleed is a term, and the selected medical record contains a value resulting in a noline translation, the line becomes invalid and is removed from the generated output.

• A potentially significant term is a term that is not derived from another term. For instance, the terms PCode(age), PCode(female or male), and PCode(pid) are examples of derived terms, i.e. their values depend on the value of another term found in the medical record, namely the pcode-term. The following terms are examples of non-derived terms: Occup, Born, Smoke. A potentially significant term has the potential of creating a significant line, i.e. if the term contains content that translates to a significant value (i.e. not noline) the line that contains it becomes significant.

• A significant term is a potentially significant term that, at generation time, is shown to have significant content (i.e. not a value that results in a noline translation). A line containing a significant term and no other invalid terms will be deemed significant at generation time.

Rules and the Generation Algorithm

The template is made up of a number of sections, each containing a number of lines, each (optionally) containing a number of terms. When producing the generated journal, each section that the user has chosen to include is processed one after the other. During the processing of a section, various rules decide whether or not the section has anything of interest, and thus whether or not it should be included at all in the output journal.

Rule 1 - If a section contains only plain text, it is included. This is so, because the creator of the template could only have produced such a section for one purpose - namely to be included if chosen. There exists no other reason to create such a section.

Rule 2 - If a section contains only plain text and derived terms, the section is included.

Rule 3 - If a section contains at least one potentially significant term, and at least one significant line is produced during generation, the section is included.

Rule 4 - If a section contains at least one potentially significant term, and no significant line was produced during generation, the section is removed.

Rule 5 - If the content between two successive titles contain no valid text, the first title is removed and the second title will begin where the first began. For instance, in the following text: "DISEASES\n\nThe patient suffers from disease.\n\nMEDICATION\n\nThe patient takes drug. Side effects that the patient recognizes are side-effects.", with the title lines "DISEASES" and "MEDICATION", if the term disease produces a no-line translation, the line containing it will become invalid. Thus, according to the rule, the line and the title containing it are removed, and the text will become "MEDICATION\n\nThe patient takes drug. Side effects that the patient recognizes are side-effects.".

Rule 6 - Removing a line that is preceeded by at least two new line characters and succeeded with at least one new line character will cause the succeeding line to begin where the removed line started.

Rule 7 - Removing a line that is preceeded by at least one new line character and succeeded with another line (not a new line character) will cause the succeeding line to begin where the removed line started.

Rule 8 - Removing a line that is preceeded by another line, and succeeded with at least one new line character will make the paragraph end with the preceeding line.

The generation algorithm:

For each section S:

Define a flag foundAtLeastOneNonDerivedChild for this section, initially false.

Define a flag foundSignificantLine for this section, initially false.

Define a set T for this section, to contain all found title lines, initially empty.

For each line L in S:

If L is a title line, add L to the set T.

Check the number of terms contained in the line:

If the line contains no terms, it is a valid but insignificant line.

If the line contains terms, but all the terms are derived, the line is also a valid but insignificant line.

If the line contains terms, and at least one of the terms is a potentially significant term, the foundAtLeastOneNonDerivedChild flag is set to true.

If the line contains at least one significant term, and no invalid terms, it is deemed as a significant line, and the foundSignificantLine flag is set to true. Also, the latest addition to the title set T is tagged for inclusion.

If the line is valid and not a title line, tag the latest addition to the title set T for inclusion.

If the line is invalid, remove it.

Remove all elements of the set T that are not tagged for inclusion.

If the foundAtLeastOneNonDerivedChild flag is still false, the section is included. If the foundAtLeastOneNonDerivedChild flag is true, but the foundSignificantLine flag is still false, the section is removed.

Remove all occurences of two or more whitespace characters in succession.

Remove all occurences of two or more line-enders (excluding newline) in succession.

Some comments:

• A section containing only plain text (no terms at all), or containing plain text and only derived terms, will be included.

An Example

In the following section-divided text, lines with a greenish tone are title lines and lines with a reddish tone will become significant due to the fact that the following terms will produce a non-noline translation: Adv-drug, Occup, Ref-in, Ref-cause, Drug, Symp-now, Time-dry, Dent-treat, HAD-A, Diag-nr, Note17. Potentially significant terms are purplish, and derived terms are blueish. The sections with a yellowish background will be removed. The result after generation (without any values for the mentioned terms above though) is showed below.

---> generation produces --->