The interpretation of missing data is a difficult issue. When a particular finding is not mentioned, it may be unknown because the value was never determined, false and not deemed a significant negative, or true but considered redundant in the context of the case. The general strategy in HFP is to handle the data by categories. If any value in a category is specified, it is assumed that all items in that category are known within a set of constraints about what findings can hide others. For example, if an EKG finding of LBBB (left bundle branch block) is specified, EKG findings of LV hypertrophy and LV strain are considered unknown because they can be obscured by LBBB and all other EKG findings, such as RV hypertrophy or long PR interval, are considered false because they would have been observed in reading the EKG. From examination of the assertions made in typical cases, this mechanism seems to work well for test results, but works less well for physical examination findings and history findings. For example, the jugular venous pressure was often noted but rarely were any other characteristics of the jugular pulse commented on, so they were treated as absent under the assumption they would have been noticed when determining the jugular pressure. However, these other findings require more careful observation than just determining the pressure. This means that the probabilities of diseases such as tricuspid regurgitation producing jugular findings appears low, while in more completely described physical exams, they might be higher. Such missing physical exam findings may be due to the summarization process or to increased reliance among physicians on test results such as echocardiography instead of the more subtle physical exam findings.
Some findings are related to one another in ways that are not readily captured by the probabilistic formalism. For example, in case 128 there is significant tachypnea (rapid breathing) but no dyspnea (difficulty catching breath) mentioned. Since tachypnea is easily and reliably measured by the observer and dyspnea is dependent on the patient's description and perceptions, the expert takes the more reliable evidence and attributes the lack of dyspnea to the process of data collection. For the program, the lack of dyspnea and lack of other evidence of pulmonary venous congestion (attributable in this case to the acuteness of the situation) ruled against high LAP as an explanation.
Symptoms such as chest pain presented a particular problem. On the program data collection menu, chest pain is a separate category, so initially if it was not marked as present or absent, the program treated it as unknown. Since it is reasonable to expect that chest pain would be mentioned if it had been present within the hours prior to the examination, the program was changed to assume that the values representing recent chest pain are false if not entered. There are other findings where such assumptions would be reasonable, such as extreme tachypnea or hypotension, but the program needs more capabilities of handling severity to take advantage of this information. In a number of cases no therapies were listed on admission and the program assumed all therapies were false. For most of these the patient was probably not taking any medications, but there were several cases where the experts suspected drug toxicity from chronic medications that were probably left out of the summary. For example, a chronic hypertensive patient with low blood pressure might have been receiving too much anti-hypertensive, even though it was not listed.