In (d-f), data are shown beginning in 2001 because the dataset for 2000 is incomplete owing to the loss of power caused by the large eruption and the subsequent lahars.
Our proposed pipeline has smaller throughput compared to state-of-the-art [7] (Table 1) because (1) we iterate the dataset for an additional pass; and (2) we have inferior computation capability with our hardware platform compared to [7].
Because these properties were observed within the dataset for the variables of interest, the use of linear regression was justified.
This database was an excellent source of datasets for this study because the datasets contained additional sample information, which was easily accessible and analyzable.
There were also large numbers of procedures that could not be analyzed, either because the datasets could not be linked (for example, due to missing patient identification details) or because there were specific data fields with incomplete data.
For a ⩾3-grade improvement in PGA and a ⩾50% improvement in NAPSI score, we separately investigated only the predictors because datasets for the complete ensemble of possible predictors were missing in >20% of patients with skin and nail lesions; thus, a selection bias could not be excluded.
It was less successful because the Lareau dataset for histidine starvation was actually rather small, too small for RRT analysis as described here.
Among the 1,451 estimated deaths, 25% were not attributed to a pathogen, mainly because the dataset did not include data for Toxoplasma spp. *Most estimates from (1); some were made as described in Methods.
Among the estimated 57,000 hospitalizations, 8% were not attributed to a pathogen, mainly because the dataset did not include data for Toxoplasma spp. *Most estimates from (1); some were made as described in Methods.
We did not include dehydration because the dataset did not contain specific codes for dehydration.
