To face the coronavirus disease-19 (COVID-19) pandemic, several countries have implemented
lockdown measures (LM), which included social distancing, quarantine, and self-isolation
to prevent virus transmission. Additionally, the current outbreak has dramatically
affected health care facilities, with consequent dynamic reorganization of health
care services, deprogramming, and contraction of nonemergency surgery procedures and
implementation of new protocols for (non-) COVID-19 patients access to health care
service.[1]
[2]
[3]
[4]
Clinicians have noted a decline in the number of patients seeking medical care for
non-COVID-19-related causes which also entailed to decreased hospital admission for
cardiovascular diseases (CVDs) such as atrial fibrillation (up to 47%)[4]
[5] and stroke (12–40%).[4]
[6] In this context, it is also possible to speculate a parallel reduction in the access
to long-term treatments for CVDs. To the best of our knowledge, the impact of COVID-19
outbreak on the management of oral anticoagulants (OACs) treatment is still unknown
and warrants a thorough evaluation. Therefore, we aimed to estimate the impact of
LM on OACs by using data from Tuscany, a region of central Italy, one of the first
countries massively involved in this emergency. We selected all adults (≥ 18 years)
with at least one dispensing of vitamin-K antagonists (VKAs) and/or direct OACs (DOACs:
dabigatran, rivaroxaban, edoxaban, and apixaban), between January 7, 2019 and September
27, 2020. New users (NUs) were those with no OACs use in the year before the first
observed prescription (washout period).
For each drug class the weekly NUs count and incidence per 10,000 inhabitants were
estimated using the adult population living in Tuscany at January 1 of each corresponding
calendar year as reference population (Source Italian Office of National Statistics).
Three periods were considered: prelockdown (before March 9, 2020), lockdown (March
9, 2020–June 14, 2020), and postlockdown (from June 15, 2020). Frequency measures
across periods were compared by using nonparametric test and relative change of mean
values. An interrupted time-series (ITS) analysis with a Poisson generalized additive
model was used to assess significant changes (p-value < 0.05) either in the level or in the slope of the time series of OACs NU among
the three different lockdown periods (DOAC and VKAs separately).[7]
[8] A level change means an abrupt effect of intervention whereas a change in slope
represents a gradual change in the estimated outcome.[6]
This study was approved by the “Agenzia Regionale di Sanità della Toscana” Internal Governance Board.
The weekly incidence of OACs NUs ([Table 1]), significantly decreased between pre- and lockdown period for both DOACs (relative
change: –36.4%) and VKAs (–50%). Conversely, the incidence of OACs significantly increased
during the postlockdown period for DOACs (+34%) but not for VKAs (+6%).
Table 1
Weekly count and incidence of new users of DOAC and VKA in Tuscany, in the three time
periods prelockdown, lockdown, and postlockdown
|
Outcome
|
DOAC new users
|
VKA new users
|
|
Lockdown
|
Relative change
|
Lockdown
|
Relative change
|
|
Pre
|
During
|
Post
|
During
Vs.
Pre (%)[a]
|
Post
Vs.
During (%)[b]
|
Pre
|
During
|
Post
|
During
Vs.
Pre (%)[a]
|
Post
Vs.
During (%)[b]
|
|
Weekly count
|
|
Mean ± SD
|
336.8 ± 56.04
|
215.5 ± 47.61
|
289.3 ± 28.92,3
|
–36.02
|
34.25
|
109.8 ± 26.8
|
59.5 ± 14.041
|
63.2 ±
9.412
|
–45.81
|
6.22
|
|
Median
Q1–Q3
|
339
298–369
|
220
173–2581
|
294
263–3102,3
|
|
|
102
92–121
|
61
48–681
|
62
57–682
|
|
|
|
Weekly incidence per 10,000
|
|
Mean ± SD
|
1.06 ± 0.18
|
0.68 ± 0.151
|
0.91 ± 0.092,3
|
–36.03
|
34.26
|
0.35 ± 0.08
|
0.19 ± 0.041
|
0.20 ± 0.032
|
–45.66
|
6.38
|
|
Median
Q1–Q3
|
1.07
0.94–1.16
|
0.69
0.55–0.811
|
0.93
0.83–0.982,3
|
|
|
0.32
0.29–0.38
|
0.19
0.15–0.221
|
0.20
0.18–0.222
|
|
|
Abbreviations: DOAC, direct oral anticoagulant; Q1, first quartile; Q3, third quartile;
SD, standard deviation; VKA, vitamin-K antagonist.
Note: p-value < 0.05 with Bonferroni correction for differences among periods: 1 - During
vs. pre; 2 - Post vs. pre; 3 - Post vs. during.
a Relative change between lockdown and prelockdown periods.
b Relative change between postlockdown and lockdown periods.
The ITS analysis depicts a significant slope change in the weekly incidence of DOACs
NUs during the lockdown period, with an initial reduction (β = –0.25, incidence ratio [IR] 0.78; 95% confidence interval [IC95%] 0.74–0.83) followed
by another slope change at 4 weeks after LM implementation (β = 0.31, IR 1.36; 1.29–1.45). Finally, 1 week after reopening we observed a slope
change (β = –0.06, IR 0.94; 0.90–0.97). The incidence of VKAs decreased in the lockdown period,
with a level change (β = –0.42, IR 0.66; 0.56–0.77) and no other significant variation until the end of
the observation period ([Fig. 1]).
Fig. 1 Effect of lockdown on incidence of direct oral anticoagulant (DOAC) (A) and vitamin-K antagonist (VKA) (B) new users in Tuscany. Points represent weekly incidence of new users of DOAC (A) or VKA (B). The gray vertical solid line is the starting date of the lockdown period, and the
gray vertical dashed line is the lockdown end date. The black solid line represents
the predicted regression model^ line with 95% confidence intervals (gray bands), while
the black dashed line represents the regression model prediction in case of no lockdown.
^Poisson generalized additive model (A), with Y = weekly count of DOAC new users and population as offset variable, f(week) as a spine function of time with 11 degrees of freedom, the holiday indicator (0 = no, 1 = yes), the lockdown week indicator (0–14) and its delayed effect (lag
4(lockdown week)) and the delayed effect of postlockdown week (lag
1(postlockdown week)). ^Poisson generalized additive model (B), with Y = weekly count of VKA new users and population as offset variable, f(week) as a spine function of time with 6 degrees of freedom, the holiday indicator (0 = no, 1 = yes) and the lockdown period indicator (0 = prelockdown,1 = after lockdown).
Both regression models had a good goodness of fit (R
2 adjusted for DOACs 0.78, for VKAs 0.89).
Our findings suggest a dramatic change in OACs use after national LM implementation.
As far as we know, this is the first study that analyses OACs use during COVID-19
outbreak since most published studies focused only on hospital admission due to CVDs.[4]
[9]
[10]
[11]
[12]
[13]
[14]
Different factors might explain the observed results. On one hand, it is possible
to assume that the delayed emergency department access due to fear of contagion might
have caused CVDs underdiagnosis in the general population; on the other hand, the
health care overload, along with the hospital adaptation/reorganization for COVID-19
cases might have limited non-COVID-19 patients' access to the health care services.
Additionally, the drop/postponement of elective surgery services during lockdown might
have caused contraction of OACs use as prophylactic treatment in perioperative phases.
DOACs have a broad range of indications, therefore it is also conceivable that the
COVID-19 countermeasures caused their abrupt reduction due to markedly reduction in
cardiology specialist visit access and cancellation of nonurgent elective surgeries.[1] Conversely, VKAs use is less influenced by reduced access to specialist care because
in Italy primary care physicians can initiate a VKAs treatment. The persistence of
VKAs underuse reduction observed in the last phase of lockdown might be explained
by the implementation of the updated cardiology guidelines which recommended not initiating
VKAs treatment during the outbreak.[3]
[15]
[16] This hypothesis is supported by postlockdown data, which indicates similar incidence
for VKAs users, but not for DOACs, as compared with that observed during the lockdown
period.
This study has potential limitations such as the lack of comparison time series that
could strengthen our results. However, we tried to limit many factors that could affect
the analysis. Changes in the population structure may bias results, but we did not
find population structure changes in the short period analyzed. We accounted for seasonality
and autocorrelation in regression model by using spline function of time and holiday
indicators. Additionally, it should be noted that the study focused only on drugs
that require continuous monitoring. Therefore, the variation in prescription patterns
can be considered less influenced by seasonality. Lastly, the Italian pharmaceutical
claim database does not include information about the indication of drug dispensing,
thus not allowing to clinically describe the reasons for drug use between periods.
However, a recent study,[17] using the same administrative database of Tuscany, reported similar reduction of
hospitalizations for several CVDs, including atrial fibrillation and stroke, during
the lockdown period compared with the same period of previous years. This may support
the hypothesis of a nondifferential underdiagnosis of OACs indications with consequent
decreased treatment initiation during the lockdown phase.
In conclusion, the observed phenomenon might result by an interplay of policies, clinical,
and social circumstances. Further studies are warranted to deeply describe this phenomenon
by considering also the second LM implementation. These findings might be useful to
reconsider the management of long-term treatments under similar exceptional circumstances.
