McElnay et al developed a model for predicting the risk of an adverse drug reaction in an elderly patient. This can help identify a patient who may require closer monitoring for an adverse event. The authors are from the Queen's University of Belfast and Antrim Area Hospital in Northern Ireland.
Patient selection: >= 65 years of age, admitted to the hospital
Parameters:
(1) prescription for antidepressants
(2) prescription for digoxin
(3) gastrointestinal problems (nausea, vomiting, diarrhea, other)
(4) serum potassium concentration
(5) patient's opinion about cause for hospitalization
(6) history of angina
(7) history of chronic obstructive pulmonary disease (COPD, or COAD for chronic obstructive airway disease)
|
Parameter
|
Finding
|
Points
|
|
prescription for antidepressants
|
no
|
0
|
|
|
yes
|
1
|
|
prescription for digoxin
|
no
|
0
|
|
|
yes
|
1
|
|
gastrointestinal problems
|
none
|
0
|
|
|
present
|
1
|
|
serum potassium
|
< 3.6 mmol/L
|
1
|
|
|
3.6 - 5.2 mmol/L
|
0
|
|
|
> 5.2 mmol/L
|
1
|
|
patient's opinion about the cause of the hospitalization
|
not related to medications
|
0
|
|
|
related to medications
|
1
|
|
history of angina
|
no
|
0
|
|
|
yes
|
1
|
|
history of COPD
|
no
|
0
|
|
|
yes
|
1
|
X =
= (1.7569 * (points for antidepressants)) + (0.6884 * (points for digoxin)) + (0.7704 * (points for GI problems)) + (0.9455 * (points for serum potassium)) + (1.4375 * (points for personal opinion)) - (1.7861 * (points for angina)) + (0.8779 * (points for COPD)) - 1.0997
probability of an adverse drug event =
= 1 / (1 + EXP((-1) * X))
NOTE: When the equation is used, values do not match the text. On page 52, it give the following relationships for different combinations of single risk factors (others negative):
(1) antidepressants only: 6 times greater
(2) digoxin only: 2 times greater
(3) GI problems only: 2.16 times greater
(4) COPD only: 2.4 times greater.
(5) abnormal potassium: 2.5 times greater.
(6) angina only: 6 times lower
(7) patient thought the drug was responsible: 4 times greater
Assuming that the error is in the equation constant and calculating backward, a constant of -5.0997 gives values very close to the text.
Performance:
• The sensitivity of the model for identifying significant adverse drug events was 41% with a specificity of 69%.
