Causes and consequences of the opioid epidemic in the Netherlands: a population-based cohort study

Over the past decade opioid use has risen globally. The causes and consequences of this increase, especially in Europe, are poorly understood. We conducted a population-based cohort study using national statistics on analgesics prescriptions, opioid poisoning hospital admissions and deaths in the Netherlands from 2013 to 2017. Pain prevalence and severity was determined by using results of 2014–2017 Health Interview Surveys. Between 2013 and 2017 the proportion of residents receiving opioid prescription rose from 4.9% to 6.0%, and the proportion of those receiving NSAIDs decreased from 15.5% to 13.7%. Self-reported pain prevalence and severity remained constant, as 44.7% of 5,119 respondents reported no pain-impeded activities-of-daily-living in 2014 (aRR, 1.00 [95% CI, 0.95–1.06] in 2017 vs 2014). Over the observation period, the incidence of opioid poisoning hospitalization and death increased from 8.6 to 12.9 per 100,000 inhabitants. The incidence of severe outcomes related to opioid use increased, as 3.9% of 1,343 hospitalized for opioid poisoning died in 2013 and 4.6% of 2,055 in 2017. We demonstrated that NSAIDs prescription decreased and opioid prescription increased in the Netherlands since 2013, without an increase in pain prevalence and severity. Consequently, the incidence of severe outcomes related to opioids increased.

concurrently an increase in oxycodone use has been noted 9 . However, it is unknown whether the increase in opioid prescription is paralleled by a similar decrease in NSAIDs prescriptions, which could offer an explanation of the opioid crisis.
Together with the increased opioid prescription, an increase in opioid related fatalities, hospitalization and death due to opioid poisoning, was observed. However, information whether hospitalization and death associated with opioid use were also boosted by an increase in illicit opioid use is currently unknown for the Netherlands. Studies from the United States have shown that wide-spread use of prescription opioids in the community preceded illegal opioid trade 10,11 . The latter introduces an additional risk for severe outcomes, such as overdose, as users of illicit opioids are not monitored 12 . Therefore we assessed, among those who were hospitalized or died due to opioid poisoning, how many had not been reimbursed for an opioid prescription, which indicates either in-hospital administration or illicit use. Furthermore, we examined different outcomes of opioid poisoning, i.e., death, prolonged hospitalization or complete recovery, in order to estimate whether the opioid crisis deepened since 2013. We aimed to explore two possible causes for increased opioid prescription in the Netherlands: increased pain prevalence and severity, and decreased NSAIDs prescription. Furthermore, we estimated consequences of increased opioid prescription such as hospital admission and death due to opioid poisoning on a population level. The causes and consequences of increased opioid prescription warrant knowledge on precautions needed to be made to prevent further increase in opioid-related fatalities.

Methods
Participants. We conducted a population-based cohort study in the Netherlands. Detailed method descriptions have previously been reported 2 . In brief, we performed analyses into prescription reimbursement data, hospitalization, and mortality data using several anonymized databases from Statistics Netherlands (CBS) covering the total population of the Netherlands between 2013 and 2017 [13][14][15][16] . Furthermore, we used data from the national Health Interview Surveys, "Gezondheidsenquête" (GE) from the years 2014-2017 (also collected by the CBS) 17 . Datasets were linked on an individual level, based on the unique anonymized identifiers.
The institutional review board of the Anesthesiology Department and Intensive Care Unit of the Leiden University Medical Center approved the study in CBS and waived participant consent.
Pain prevalence and severity. Information on pain perception was included in the GE surveys from 2014 to 2017. The GE is an annual national survey that covers health-related lifestyle choices of Dutch residents 17 . Since 2013, the GE surveys have been conducted as part of the European Health Interview Survey (EHIS) and direct comparisons between the yearly GEs can be made for the years 2014-2017 18,19 . The survey is sent to a random subset of the population of around 15,000 individuals each year, with response rates between 60 and 65% 17 . Over the four-year period around 38,000 individuals participated in the survey.
To explore the prevalence and severity of pain in the population, we investigated pain-impeded activities of daily living (ADL). Pain-impeded ADL was defined as the level of performance of daily activities (including outdoor and household chores) hindered by pain in the past four weeks 17 . Pain-impeded ADL was asked on a five-point Likert scale, with the categories: "not at all", "somewhat", "moderate", "much", and "extreme". Due to a low number of respondents in the "much", and "extreme" categories, these two groups were merged into the category "much and extreme". Analgesic prescription. Prescription reimbursement data were collected for all Dutch residents entitled to pharmaceutical care, i.e., those with health insurance, which is 99.9% of all Dutch residents 20 . The Dutch Health Care Institute (ZINL) provides these data to the CBS. Analgesics dispensed from outpatient, community pharmacies, and in residential homes for elderly are collected in the national reimbursement database, whereas medicines dispensed in hospitals and nursing homes are not 14 .
Hospital admissions and deaths related to opioid use. The Dutch Hospital Database contains information about all-cause hospital admissions and the Register of Causes of Death records all-causes of death. Each hospital admission record contains the date of hospital inpatient and outpatient encounters, the discharge date, and the discharge diagnoses 15,16 . Diagnoses and deaths are coded within the CBS according to the International Statistical Classification of Diseases and Related Health Problems (ICD, 10th revision) of the WHO 23 .
Hospital admissions and deaths registered as due to opioid-related disorders, adverse events of opioid use and opioid overdose were defined as opioid poisoning (see Supplement online) 24  www.nature.com/scientificreports/ are presented as counts and as a proportion of the total population per calendar year. Hospital admissions, deaths, and also prescriptions are presented as number per 100,000 inhabitants per calendar year and as risk ratios with 95% confidence interval (CI) compared with the reference year (2013). Risk ratios were adjusted for age and sex (aRR) with direct standardization; age was divided in 5 categories (0-15, 15-25, 25-45, 45-65, ≥ 65 years), and further stratified by sex with weights from the total Dutch population of 2013. Similar analyses were performed on the 2014-2017 GE cohorts where the selected reference group was the 2014 GE cohort, whence the weights for standardization analysis were selected. To investigate whether GE surveys were a valid representation of the total population, we linked GEs data with the prescription datafiles of the same calendar year, and performed frequency analysis into opioid and NSAIDs prescription among respondents of the GEs.
Missing data. Individuals with missing data in the pain-impeded ADL variable in the GEs were dropped from the analysis (n = 4,569, 46.5% in 2017). In 2013, 8 (0.6%) hospitalization cases due to opioid poisoning had missing information about residence before and destination after hospitalization. In 2014, 31 (2.2%) cases had missing information about residence before hospitalization due to opioid poisoning, and 29 (2.0%) cases had missing information about destination after hospital admission due to opioid poisoning. After 2014, there were no missing data. There were no missing data for the total population characteristics and no individuals were lost in the linkage process.
Hospital admissions and deaths related to opioid use. To study the impact of increased opioid prescription we performed several analyses. First, we estimated opioid poisoning per calendar year (either leading to hospitalization or death). Then we stratified opioid poisoning cases in two categories: whether individuals received opioid prescription or none in the same calendar year. This provides insight in severe outcomes-defined as death, or consequent transfer to another health facility after being hospitalized for opioid poisoning-related to prescription opioid use versus in-hospital or illicit opioid use. Second, we explored residence status prior to hospitalization. Those who were transferred from another health facility to a hospital for an opioid poisoning were considered poisoned whilst being hospitalized, which provides information about in-hospital opioid use. Third, we assessed the severity of opioid poisoning per calendar year by following-up patients after their hospitalization. We classified three main outcomes of opioid poisoning: returning home, prolonged institutionalization, or death. Those who were able to return to their own living environment were considered to have experienced a milder form of poisoning. Those who were transferred to another health facility were considered prolongedly institutionalized due to a more severe poisoning. Individuals who were transferred to a psychiatric hospital were considered having an opioid addiction. Patients who died after being hospitalized for opioid poisoning were considered having experienced an opioid overdose. The STROBE statement checklist for cohort studies is included in the Supplement online. All statistical analyses were performed with SPSS for Windows, release 24.0 (SPSS, Chicago, IL, USA). Figures were created with R studio (A Language and Environment for Statistical Computing, R Core Team, R Foundation for Statistical Computing, Vienna, Austria, https ://www.R-proje ct.org), using R package ggplot2 version 3.2.1 25 .

Results
Description of the studied populations. For the evaluation of analgesic prescription practice and consequences of changes thereof we studied 2 populations: the total Dutch population between 2013 and 2017, and the GE survey participants between 2014 and 2017.
For the total Dutch population, among the 16 Is there an increase in pain prevalence and severity? The missing proportion for pain-impeded ADL in GE cohorts was approximately 46% and the level of missingness did not change between 2014 and 2017 ( Table 1)

Discussion
We previously reported an increase in opioid prescription and related fatalities in the Netherlands from 2013 to 2017 2 . In the present study, based on national statistics, and annual population-wide national surveys we further elaborated on causes and consequences of the increase in opioid use. We found a shift from NSAID prescription to opioid prescription, without an overall increase in need for pain treatment. Over the four-year period,  www.nature.com/scientificreports/ opioid prescriptions increased by 20%, whereas opioid poising increased by nearly 50%. The increase in opioid prescriptions was mainly due to a large increase in the use of natural opioids. The severity of opioid poisoning also increased, since there were more opioid-related deaths among those admitted with opioid poisoning, and the number of those who were consequently transferred to the other care facilities had risen. The increase in opioid prescriptions was mainly due to an increase in natural opioid prescriptions (N02AA), namely, morphine, hydromorphone, and especially oxycodone, and not in synthetic opioid prescriptions (N02AZ). At the same time a decrease in NSAIDs prescription rate has been observed, although still a large proportion of residents (13.7% of the total population) received an NSAIDs prescription in 2017. Of note, NSAIDs are not only prescribed, but are also sold over the counter in the Netherlands [26][27][28] . We demonstrated, that a recent increase in opioid prescription cannot be sufficiently explained by increase in prevalence and severity of pain, but that changed analgesic prescription practice, defined as a shift from NSAIDs to opioids prescription, is the most probable reason for opioid epidemic in the Netherlands. The change in analgesic prescription practice subsequently stemmed from a concurrent introduction of oxycodone by revised Dutch pain treatment guidelines 8 , and restriction of NSAIDs use due to their common adverse events by the scientific community 29,30 .
We considered suspicion bias in the observed increase of opioid-related fatalities 31 . When suspicion bias would have been the explanation for the increased rate of opioid-related fatalities, an increase in hospital admission and death due to opioid poisoning would have been restricted to those receiving an opioid prescription in the same calendar year. However, the increase also included individuals who had not received opioid prescriptions. These individuals most probably had acquired the drugs illegally, since we found a few in-hospital poisoning (3.9% of all opioid poisonings in 2017). Furthermore, those who had not received an opioid prescription were considerably younger (mean age difference was ten years) compared with those receiving opioid prescriptions and mostly males (65% of all opioid poisoning cases in those not having prescription), which further supports our finding that among those who had not received an opioid prescription, opioid poisoning occurred due to illicit opioid use. These observations render suspicion bias as an explanation for the observed increased rate of opioid poisoning unlikely. Furthermore, this finding is consistent with reports on the opioid epidemic from the United States that showed that widespread opioid use leads to widespread opioid addiction (either prescription or illegal use), with gradual increase in severity of consequences (fatal or non-fatal opioid poisoning) 32 .
Opioid use and opioid overdose deaths are increasing in most countries in the European union [33][34][35][36][37] . However, the situation of pharmacologic pain relief in the Netherlands is somewhat different compared with other European countries. For instance, Danish and British pain guidelines advocate NSAIDs as first line treatment, and are far more stringent in opioid prescription compared to the Dutch pain guideline 38,39 . Furthermore, a decrease in prescription opioid use has been noted since 2016 in the United Kingdom 40 . Specifically, in the United Kingdom Table 3. Residence before and destination after hospitalization for opioid poisoning in the Netherlands, from 2013 to 2017. aRR, adjusted risk ratio; RR, risk ratio; CI, confidence interval. *Adjusted for age and sex with direct standardization; 2013 cohort was selected as a reference. a Academic, general, categorical, psychiatric. b Rehabilitation institution, nursing/residential home, other institutions, hospital abroad, born in this hospital, origin unknown. c Academic, general, categorical. d Rehabilitation institution, nursing/residential home, other institutions, hospital abroad, hospice, destination unknown. Opioid poisoning cases were identified in the hospitalization dataset by the ICD-10CM codes reported in the Supplement online.  www.nature.com/scientificreports/ and in Denmark, tramadol is the most frequently prescribed opioid, and not natural opioids, such as oxycodone (the opioid that is advocated by Dutch pain guidelines in favour of NSAIDs) 41,42 . This reinforces our finding that a changed analgesic prescription practice, which was preceded by pain guidelines, is indeed responsible for the recent opioid epidemic in the Netherlands. In addition, we showed that, over time, more patients in the Netherlands suffered from prescription opioid poisoning than from illegal opioid poisoning, while in the United Kingdom the far majority of patients have opioid poisoning related with heroin use, i.e., illegal opioid poisoning 33 , further supporting the evidence that the opioid epidemic in the Netherlands is for a large part iatrogenic. This research has some methodological issues that warrant commenting. First, to estimate pain prevalence and severity we used results of national health surveys. All other outcomes were identified in the Dutch national statistics. Second, the question related to pain prevalence was missing in 46% of participants. As we consider it unlikely that the missingness was at random, while at the same time the amount of missingness was large, we decided not to impute missing values. However, we consider it unlikely that the reason for missingness changed over the observation period and the level of missingness did not change between 2014 and 2017. Third, nonresponse bias cannot be excluded as survey response rates ranged between 60 to 65%, depending on the year. Although, this is an acceptable response rate of questionnaires in social sciences 43 , it may have affected our results. However, population characteristics of survey participants were similar to the total Dutch population from 2014 to 2017 (see the Supplementary Table S1 online), as well as participants were randomly sampled. Therefore, it is unlikely that the results of the national survey are not representative of the whole population. Nevertheless, the missing data may render the absolute numbers of individuals reporting pain inaccurate, where an overestimate seems most likely. Additionally, the frequency of opioid and NSAIDs prescription among the respondents of surveys was similar to the frequency of opioid and NSAIDs prescription in the total Dutch population (see the  Supplementary Table S4 online). Fourth, we did not have the detailed prescription information to enable us to identify individual active substances, dosing, and pharmaceutical dosage forms. Fifth, we only performed research into opioid and NSAIDs prescriptions, but not into other analgesic agents, such as antidepressants and antiepileptic agents, as those are mostly used for neuropathic pain 44 . Sixth, NSAIDs are also available as an over the counter medication, and the information about the proportion of the population exposed to them is unknown. Although, the NSAIDs prescription decreased between 2013 and 2017 in the Netherlands, that does not necessarily imply that the total exposure to NSAIDs in the total population decreased, because NSAIDs are also available over the counter. Lastly, hospital diagnosis and deaths were ICD-10 coded by the CBS and the accuracy of these codes is not known.
In conclusion, the opioid prescription rate is increasing in the Netherlands, without an increase in pain prevalence and severity. This increase is mainly related to natural opioid use, while at the same time NSAIDs prescription is decreasing. This shift in analgesic prescription practice was accompanied by the increase and the worsening of opioid toxicity, which was related to prescription opioids and increased illicit use of opioids. The changed analgesic pharmacotherapy strategy in the Netherlands has potentially exposed more individuals to toxic effects of opioid use, and without taking any measures to prevent further deterioration of pain management, it doesn't seem that rate of opioid-related fatalities will decline any time soon.

Data availability
Data obtained in all analyses cannot be shared with third parties as Statistics Netherlands does not permit this.