Monte Carlo method for the evaluation of symptom association

Abstract

Gastroesophageal monitoring is limited to 96 hours by the current technology. This work presents a computational model to investigate symptom association in gastroesophageal reflux disease with larger data samples proving important deficiencies of the current methodology that must be taking into account in clinical evaluation. A computational model based on Monte Carlo analysis was implemented to simulate patients with known statistical characteristics Thus, sets of 2000 10-day-long recordings were simulated and analyzed using the symptom index (SI), the symptom sensitivity index (SSI), and the symptom association probability (SAP). Afterwards, linear regression was applied to define the dependency of these indexes with the number of reflux, the number of symptoms, the duration of the monitoring, and the probability of association. All the indexes were biased estimators of symptom association and therefore they do not consider the effect of chance: when symptom and reflux were completely uncorrelated, the values of the indexes under study were greater than zero. On the other hand, longer recording reduced variability in the estimation of the SI and the SSI while increasing the value of the SAP. Furthermore, if the number of symptoms remains below one-tenth of the number of reflux episodes, it is not possible to achieve a positive value of the SSI. A limitation of this computational model is that it does not consider feeding and sleeping periods, differences between reflux episodes or causation. However, the conclusions are not affected by these limitations. These facts represent important limitations in symptom association analysis, and therefore, invasive treatments must not be considered based on the value of these indexes only until a new methodology provides a more reliable assessment.

The binomial symptom index: toward an optimal method for the evaluation of symptom association in gastroesophageal reflux

“…Reflux and symptom occurrences can be considered as two independent arrival processes . In some occasions, there may be correlation between both events and therefore a positive association.…”

“…To test the BSI, a Monte Carlo model was implemented according to scientific literature . This model assumes that reflux and symptom arrivals are distributed following a Poisson process.…”

“…In this way, different indices were suggested to determine the degree of association between reflux and symptom events: the symptom index (SI), the symptom sensitivity index (SSI), and the symptom association probability (SAP). The SI does not consider the total number of reflux episodes, the SSI failed for not considering the total number of symptoms, and neither takes into account the effect of chance . The SAP appeared to overcome these problems although it has recently shown important deficiencies: it changes abruptly over the width of the window of association and increases with the length of monitoring .…”

“…In some studies, a small number of patients represent a problem when looking for representative sample of the population. On the other hand, in silico methods have recently proved to be of relevant utility when assessing the goodness of SAA because they provide enough data to draw solid conclusions . Monte Carlo simulations allow generating unlimited amounts of virtual patients with well known statistical features.…”

“…This index considers all the parameters involved in symptom association and can provide simultaneously a measure of probability and strength of association (effect of size) when the cut‐off value is defined appropriately. Its goodness was tested computationally against different parameters by means of Monte Carlo simulations as described in the scientific literature …”