Tisdale Score for QT prolongation

Clinical background

Drug-induced QTc prolongation is common among inpatients and can lead to torsade de pointes (TdP), a potentially lethal arrhythmia. In cardiac care units, up to 28% of patients already show QTc prolongation on admission, and about a third of these subsequently receive further QTc-prolonging drugs [1]. The risk of TdP rises stepwise with the QTc interval and increases markedly once the QTc exceeds 500 ms [1]. Since many inpatients carry several concurrent risk factors such as electrolyte disturbances, impaired left ventricular function and acute cardiovascular disease, a way of quantifying the combined risk is needed in order to prioritise monitoring and to avoid or substitute hazardous drugs. The Tisdale risk score for QT prolongation was developed for precisely this purpose: to identify, using readily available clinical variables, which inpatients are at highest risk of developing QTc prolongation during their stay and should therefore receive more intensive ECG monitoring or alternative drug choices [1].

Calculating the Tisdale risk score

The score is a weighted sum of nine clinical variables, the weights being based on the log odds ratios for each independent risk factor in a multivariable logistic regression model [1]. The scoring follows the principle that a log OR ≤ 0.44 scores 1 point, 0.45 to 0.94 scores 2 points and ≥ 0.95 scores 3 points, with an additional allocation for each further concurrent QTc-prolonging drug.

$$\text{Tisdale} = 1 \cdot \mathbb{1}$${\text{age}\geq 68} + 1 \cdot \mathbb{1}{\text{female}} + 1 \cdot \mathbb{1}{\text{loop diuretic}} + 2 \cdot \mathbb{1}{K^+\leq 3{.}5} + 2 \cdot \mathbb{1}{\text{QTc}\geq 450\text{ ms}} + 2 \cdot \mathbb{1}{\text{acute myocardial infarction}} + 3 \cdot \mathbb{1}{\text{sepsis}} + 3 \cdot \mathbb{1}{\text{heart failure}} + 3 \cdot n_{\text{QT drugs}}

where $n_{\text{QT drugs}}$ is 0 with no QTc-prolonging drug, 1 with one, and 2 with two or more (giving 6 points in total for the drug variable). The range is 0 to 21.

The derivation cohort consisted of 900 consecutive patients admitted to cardiac care units (two 28-bed units) at Indiana University Health Methodist Hospital, a university-affiliated tertiary hospital in Indianapolis, USA, between September 2008 and March 2009 [1]. QTc prolongation was defined as a QTc ≥ 500 ms or an increase of ≥ 60 ms from the admission value. Its incidence was 30.7% in the derivation group and 30.0% in the validation group (n = 300, admitted between April and June 2009 to the same units) [1].

Interpretation in practice

The score ranges divide patients into low, moderate and high risk according to the following bands, with the incidence of QTc prolongation in the validation cohort [1]:

A score of 11 or more gave, in the validation cohort, a sensitivity of 0.74 and a specificity of 0.77 for predicting QTc prolongation, with a positive predictive value of 0.79 and a negative predictive value of 0.76 [1]. The main strength of the score lies in identifying patients whose risk is low enough that no specific measures are needed, rather than in classifying every individual risk precisely.

When the score was implemented as a clinical decision support system (CDSS) in the same cardiac care units, the incidence of QTc prolongation fell significantly: the adjusted odds ratio was 0.65 (95% CI 0.56 to 0.89, P < 0.0001) compared with before implementation [2]. The system also reduced the prescribing of non-cardiological QTc-prolonging drugs such as fluoroquinolones and intravenous haloperidol (adjusted OR 0.79, 95% CI 0.63 to 0.91, P = 0.03) [2].

Validation and performance

The derivation study achieved a c-statistic of 0.823 in the derivation cohort, indicating good discrimination [1]. The validation cohort (n = 300, same unit) showed corresponding results, with incidences of 15%, 37% and 73% for low, moderate and high risk [1].

An external validation in a medical intensive care unit (ICU) in British Columbia, Canada, comprising 264 patients, showed high sensitivity but low specificity: 97% (95% CI 91 to 99%) and 16% (95% CI 11 to 23%) respectively when low risk was set against moderate and high risk combined [3]. The negative likelihood ratio was 0.20 (95% CI 0.06 to 0.65), meaning that a low score reliably rules out risk [3]. The low specificity does, however, mean that many patients are incorrectly flagged as being at risk, which can lead to unnecessary monitoring.

A retrospective study across 28 hospitals in the western USA with 92,383 inpatients found a strong association between the score and mortality: the adjusted OR for death was 4.80 (95% CI 4.42 to 5.21) at moderate risk and 11.51 (95% CI 10.23 to 12.94) at high risk compared with low risk [4]. Length of stay increased by 0.7 days per one-point increase in score (P < 0.0001) [4]. This study used a modified version of the score and examined mortality rather than QTc prolongation as the outcome, but the results suggest that the score has prognostic value beyond its original endpoint.

In a population of 178 hospitalised non-intensive care COVID-19 patients, by contrast, the score performed poorly: the AUC was only 0.60 (95% CI 0.46 to 0.75), and at a threshold of 7 points sensitivity was 85.7% but specificity only 7.6% [5]. The authors concluded that the score is not useful for stratifying non-critical COVID-19 patients, illustrating that the instrument is tied to the population in which it was developed.

Limitations

The Tisdale risk score was derived exclusively in cardiac care units and their step-down units at a single tertiary hospital [1]. Its generalisability to other types of ward, particularly general medical wards and non-cardiological intensive care units, is incomplete. The external validation in a medical ICU showed low specificity [3], and the COVID-19 study found that the score was not useful in a different patient population [5].

Serum calcium and magnesium are not part of the score, even though these electrolytes affect the QTc. Magnesium was not measured routinely in the derivation cohort (in only about 38% of patients) and was therefore excluded [2]. Patients with a paced rhythm were excluded from the derivation because the QT interval cannot be measured reliably in paced rhythms [1], and the score therefore does not apply to these patients.

The score does not replace clinical judgement or serial ECG monitoring when a QTc-prolonging drug is started. It should be seen as a screening tool that can be integrated into clinical decision support systems to direct attention to the patients in whom further measures are justified, not as the sole basis for withholding or discontinuing specific drugs.

References

1. Tisdale JE, Jaynes HA, Kingery JR et al. Development and validation of a risk score to predict QT interval prolongation in hospitalized patients. Circ Cardiovasc Qual Outcomes. 2013;6(4):479-87. PMID: 23716032
2. Tisdale JE, Jaynes HA, Kingery JR et al. Effectiveness of a clinical decision support system for reducing the risk of QT interval prolongation in hospitalized patients. Circ Cardiovasc Qual Outcomes. 2014;7(3):381-90. PMID: 24803473
3. Su K, McGloin R, Gellatly RM. Predictive validity of a QT(c) interval prolongation risk score in the intensive care unit. Pharmacotherapy. 2020;40(6):492-499. PMID: 32259316
4. Tan MS, Heise CW, Gallo T et al. Relationship between a risk score for QT interval prolongation and mortality across rural and urban inpatient facilities. J Electrocardiol. 2023;77:4-9. PMID: 36527915
5. Zhao W, Gandhi N, Affas S et al. Predicting QT interval prolongation in patients diagnosed with the 2019 novel coronavirus infection. Ann Noninvasive Electrocardiol. 2021;26(5):e12853. PMID: 33963634