Review of article: “Statin Use and Risk of Cerebral Aneurysm Rupture: A Hospital-based Case-control Study in Japan,” Y. Yoshimura, Y. Murakami, et al.

Design and Data Collection

The presented study is a case-control study design, meaning it is based on measuring the associations between statin administration and the likely rupture of a cerebral aneurysm in purposefully selected indicator and control populations. The foundation of the study grew from theoretical components tied to animal models used in the past. The underlying conditions of interest are important but not commonly seen in humans, making it difficult to know the best medical measurements and, if possible, prescriptive interventions for unruptured neurological disorders, which generally now can be dealt with by way of surgical interventions.

The written assessment of the study indicated limited evidence of human studies showing the effect of statin drugs in reducing the risk of aneurysmal ruptures. The study model used in this investigation (case-control) is readily suitable to study uncommon diseases with low overall prevalence rates. However, given the impact such a condition could have on a population segment thought to have many years of productivity – and because of promising animal studies – the study was deemed beneficial as an initial step toward learning more about the drug’s beneficial indicators. The primary endpoint of focus was the likelihood of the rupture of an aneurism and the protective effect of statins on reducing that outcome.

According to the authors (Yoshimura, Murakami, et al), their results showed that use of any statin “was inversely associated with rupture risk in unruptured aneurysm patients.” This finding was qualified because of a number of limitations associated with their model and techniques. Our recommendation seeks to address some of these concerns.

The CASE subjects for this study were 117 patients of unruptured cerebral aneurysm enrolled from 15 hospitals in Japan. Several hospitals were utilized from each of the localities of Shiga (8 hospitals), Kyoto (5 hospitals), and Tokyo (2 hospitals). Their corresponding CONTROL group total of 304 patients with cerebral aneurysm came from the same institutions. The names of the hospitals were not specified. The full population of the study was enrolled in a hospital and fell within certain age parameters, mean age at presentation of 50. Their method made use of a convenience sample instead of relying on other random selection sampling approaches. It should be noted that the mean age of 50 is true even though the reach to capture the desired population allowed for patients as young as 20. The majority of participants were older and reflected diverse demographics, given the small population size tied to this condition.

The particular case participants were selected using direct criteria centering on age and the nature of the bleeding. In addition, the investigators sought basically to determine which factors might contribute to the end point (rupture), and the impact on those factors with statin usage. A number of variables were assessed and comparative data collected to determine what likely factors could be associated with statin usage. The existence of a subarachnoid hemorrhage (SAH) with bleeding was the baseline for cases; control conditions had not ruptured. Those with SAH resulting from trauma, dissecting aneurysms, infected aneurysms, autoimmune diseases, and familial diseases were excluded to limited compounding complexities. Case patients were in the hospital from April 2009 to March 2011; controls were extracted from records in April 2006.

Control participants generally came from a wider population group that appeared to have the condition known over a longer period of time, which could impact the Odd Ratio in either direction. Among the case group there were 41 males (sample size=117), while the control group had 200 (sample=304), making the proportions similar. The methods used for obtaining specific medical information on participants was not clear, though some indicators were developed for rating collected information (like putting alcohol and smoking use into general categories, never, past, current). All selections included the verification of radiological images. In general, some of these factors likely result in a conservative bias toward the control group. Control patients may be healthier overall, pushing the Odd Ratio more toward the null hypothesis. This is a common concern with studies of this nature. If this bias is true, it may be harder to prove what effect the statins had.

The largest and most critical bias factor, though, is likely connected to the lack of randomness. One conceivable way around this could be by way of using varies matching capacities. Mismatched assessments can bring about other biases, especially given that case and control patients came from several hospital locations. Each hospital has different guidelines and standard practices. Patient and provider recall biases in this regard could cause misclassifications of the disease status and again tend to favor the null hypothesis. Using other equalizing methods seemed most appropriate for aligning the results with other studies and impacts of the statins. The investigators sought to compensate for this through statistic adjustments and data softening.

Data Analysis and Results

The size and therefore the power of the sample populations were basically not discussed. It was assumed, with reasonable certainty, that the study had an adequate representation of the target groups. A preliminary review suggests the numbers are more than sufficient, but for other reasons it appears that a larger number of statin users in the case group would be preferred. The alternative model we propose seeks to deal with this.

The following table reflects the contingent factors for both study populations in a 2x2 presentation. It is likely that the numbers are adequate for reasonable statistical assessment. However, other issues can become important in reviewing the results. For example, the fact that the sampling uses more than one site can introduce layers of selection bias. The different hospitals have different record-keeping and patient monitoring techniques. These concerns can amplify population worries.

| Ruptured aneurysm | un-ruptured aneurysm | statin | 11 | 79 | no statin | 106 | 225 |

The investigators took reasonable steps to ensure competent and reliable data. A table of the confounders that they adjusted for is included. (Table 2 of study.) It appears that they were able to weight or soften the data across different datasets, improving the reliability of comparisons. They used multivariable logistical regression and chi square to judge the strength of relationships. There were no major causal associations evident. Issues with temporality and statin type and dosage seem connected to these findings. It is difficult to tell though since there are not many related studies to balance their concerns with. Those that do exist with a similar model were conducted on rats or mice, weakening the biological plausibility of comparisons.

A key area of greater concern relates to the dosage and type of statin used by participants. The investigators note this issue in their assessment and while looking forward to future studies. Various studies indicate the importance of these factors. For example, the control group is supposed to be healthier in coming to the hospital settings. Without confidence in their dosage and adherence, the confidence in this assessment is harder to ensure. Lacking information about these factors can lead to undifferential misclassifications, driving the finding more toward the null hypothesis. The fact that the results still affirmed a link between the use of statins and no ruptures, this may not be a bad result.

The methods used to gather data were clearly explained in the article. Adjusting for the complexities of having multiple hospital sites was addressed well, and some adjustments were made to accommodate this element of concern. In gathering patients for the case and the controls from many sites, the authors likely have overcome negatives lost by not randomizing the study, as did including core demographic (age, gender) factors and at least considering such medical indicators as size of the aneurysm and history of hypertension.

The exclusion of some highly prejudicial factors regarding the patients were good to. The exclusion of controls with previous SAH, for example, critically differentiated the controls from active cases. Adjustments and stabilizing factors such as alcohol use and smoking and hypertension helped the researchers deal with the results tied to these considerations. In a similar way, by anticipating cholesterol levels the authors limited an important characteristic that could be either confounding or otherwise an effect modifier.

Interpretation and Discussion

Overall, the results reasonable support the researcher’s hypotheses to a certain degree. I basically agree with their results and their assessments. I also believe they did a good job of exploring the limitations of their work as a basis for future studies. Their efforts to balance out confounding elements was reasoned and reasonable, using good logic to adapt their analysis.

One of the most significant biases or limitations was their inability to monitor the effect of statins because they did not measure lipid profiles before and after the drugs were administered. They mention this specifically in their conclusions. In a related consideration, they did not measure variations on the size of the aneurysms before, during and after administration of the drug or other clinical activities.

The problems I will address in improving the study center on patient health misclassifications related to enrolling subjects from multiple hospitals, and greater validation of statin use by controls. These factors are mostly systemic flaws tied to considerations relating to data collection across the hospital sites. Similarly, the investigators included participants across a wide array of years, particularly for the controls. This reach can be important in considering advances made in the use and testing of statins. New areas of interest, such as the impact of statins on weight, obesity, etc., could be lost in this way. Whatever statistical compensations were made can still reduce the generalizability of the findings to other hospital settings.

Proposed Methodological Improvements

To deal with the concerns, I propose making changes in the study in regards to where the study is conducted, how data is captured and recorded, and in the ways that biomarkers can be used to enhance the results. These changes seem well aligned with advancing technologies and can make the results stronger as statins are increasingly used by patients at risk.

NEW RESEARCH PLAN

Section A: Specific Aims
Cerebral aneurysm, with or without subarachnoid hemorrhage (SAH), is a relevant health problem associated with serious complications for people identified early on with possible condition indicators.1 Currently, there is no clear guideline in clinical practice covering this issue. There is a growing collection of literature that offers some guidance for medical practitioners, yet there are no best treatment modalities published other than surgical clipping or coiling. The efficacy of coiling on the long term is unsettled.
Part of the reason for this lack of progress has to do with the frequency of the condition. The prevalence of intracranial aneurysms is 2.3% (95% CI, 1.7-3.1%); most of these aneurysms are small and located in the anterior circulation. Risk factors are age, female gender, smoking, hypertension, excessive use of alcohol, having one or more affected relatives with SAH and autosomal dominant polycystic kidney disease. 2 The overall risk of rupture found in follow-up studies is around 1% per year. 2
Between 3.6 and 6% of the population harbor an unruptured intracranial aneurysm. Risk of rupture is related to aneurysm site and size and whether or not the patient has already had a subarachnoid hemorrhage (SAH) from another aneurysm. In ISUIA 2, the rupture rate for anterior circulation aneurysms <7mm was 0% per year in patients with no prior SAH, and 0.3% per year in patients with previous SAH; 7-12mm aneurysms, 0.5% per year (both groups); 13-24mm aneurysms, 3% per year; and giant aneurysms 8% per year. Rupture rate for posterior circulation aneurysms is higher at all sizes: <7mm was 0.5% per year in subjects with no prior SAH, 0.7% in those with prior SAH; 7-12mm, 3% per year; 13-24mm, 3.7% per year; and giant aneurysms, 10% per year. 3

Statin is well known as a class of one lipid-lowering agent most commonly used to lower cholesterol level in a treatment of atherosclerosis. Statin also has multiple mechanisms of action including antioxidant, anti-inflammatory and profibrinolytic effects. Statin was studied in animal models and it showed its effect in reducing the incidence of aneurysm formation and it has a vascular pleotropic effects through attenuation of intravascular inflammation. And improved the endothelial function through up regulation of nitric oxide. 4

Still, there seems to be growing confidence that a noticeable part of the population could experience related conditions. Poor health and eating, obesity, a technological society, etc., makes it possible the condition could expand if aneurysms are related, or if their rupture is. Early detection and the size of the weakness of the circulation system, for example, could bring about more cases, just as technology could show aneurysms earlier in preventative care. Knowing what statins do can become more important under these conditions.

My study seeks to anticipate this and strengthen the validity of the original study. To do that, I have identified specific aims and the reasons why they are important.

Specific Aim 1: To study the protective effect of statin use on cerebral aneurysms, and the associations between their use and subsequent ruptures by refining a case-control investigatory model.

A. In a hospital-based study, I will measure the positive association between statin use by case and control groups. The case group will be patients who are hospitalized with an identified aneurysm that has ruptured. The second, control group will be similar patients who have the condition but without rupture. This approach effectively mirrors the efforts of Yoshimura, Murakami, et al., except it expands and refines it to directly lessen some of the identified weaknesses. Participants will be approached because of their medical records, with a focus on those receiving the same types of clinical care and radiological investigation of progress. We will characterize the core demographic features of both groups and utilize data adjustments to limit confounding influences.

Specific Aim 2: To determine the role of statin anti-inflammatory actions, and compare condition measurements for effects associated with or without ruptures.

Adding the element of measuring the aneurysmal size as a baseline to the study will help to inform the drug’s impact whole facilitating the interpretation of related biomarkers (e.g., nitric oxcide) that vary from baseline.

Specific Aim 3: To examine the correlation between the cerebral aneurysm rupture potential and statin dosage in order to help clarify it as a neuroprotective agent for unruptured conditions.

A. Null Hypothesis: There will be no correlation between simvastatin at 20 mg dosage administration among the exposed subjects verses those who do not use statins. The type and dosage size was selected because of its ready use and known reactive patterns. The results of this focus will provide guidance in regard to the impact potential of other types and dosages.

Section B: Research Design and Approach

Disease, design and population base

The natural history and progression of unruptured cerebral aneurysms is still matter of investigation. The prevalence of unruptured aneurysms is around 2%; most of these aneurysms are small and located in the anterior circulation.

Due to such low prevalence rate the case control based study is efficient for rare disease. Case-control study design is cheap and less time consuming. However, with this type of study we can’t measure the incidence or absolute risk, as a cohort study might be superior in demonstrating. Because the numbers are low, the case study seems preferable. The incidence of ruptured cerebral aneurysm is about 10 in every 100,000 person per year (about 30,000 individual per year in US) most commonly in people between ages 30-60 years.5

The population-based will continue to be patients (case and control) in a hospital setting. They will be attached to a neurology clinic. The case subjects with have a subarachnoid hemorrhage due to ruptured cerebral aneurysm, while the control are disease but having no bleeding. All the participants will be enrolled in this study at the same period of time. The control should be selected randomly as well as the cases and thus we will try to reduce the selection bias.

Subject acquisition and data collection points

In general, the key strengths of this model center on the fact that we will collect the participants from more than 400 hospitals at 60 sites across the United States. This is a large number offering access to a diverse group. The findings of cerebral aneurysm will be confirmed by a comprehensive standardized method, with clear criteria of positivity of the subarachnoid hemorrhage with no significant neurological deficit in order to avoid certain complicating issues.

Part of the added value of my approach rests on the ways that it will explore changes how being implemented at this time to transfer medical records electronically. These advances can ensure more accurate data gathering. I believe we can utilize this emerging trend to compensate for not randomizing or matching patients. When possible case and control populations will be identified in like ways from each hospital to minimize influences from medical and health system variations. All data will be collected from patient medical records in consultation with their clinical providers using surveillance or index measures maintained in their charts.

The controls are patients with unruptured cerebral aneurysm of small size aneurysm (less than 7 mm in size) confirmed by MRI. The overall risk of rupture found in follow-up studies is around 1% per year. Size is the most important risk factor. Other risk factors are the site of the aneurysm (higher risk for posterior circulation aneurysms), age, female gender, population (higher risks in Finland and Japan) and, probably also, smoking. 2

The period of collection will be from January 2010 to December of 2013. The dates ensure that participants, if on statins, are likely using recent versions of recommended dosages and routines. The cerebral aneurysm is somewhat more common in adults and slightly more common in women than men; therefore cases and control should both be between ages 30-60 years of both gender with consideration to risk factors such as alcohol consumption, drug abuse (particularly cocaine), smoking and oral contraceptive in women as the estrogen has direct relationship in the mechanism of cerebral aneurysm.

The subjects in both case and control categories will consent to be in this study. We are expecting, using known statistics of incidence, to secure up to 1000 patients under the control group and approximately 300 participants who qualify as case patients (given the know low rates of the condition). As in any medical research study of this nature, it is possible that patients of either the case or control study may die or withdraw.6 We believe the target numbers are sufficient to accommodate this, but we will report how many cases were eligible and met the initial criteria for inclusion (or exclusion) and assess the results and how this will affect the selection bias. All subjects will be expected to remain involved across the study period and for one year of follow-up.

In summary, the following are summaries of the critical inclusion and exclusion factors:

INCLUSION:

* Patients between the ages of 30-60 who have ruptured (case) or unruptured (control) cerebral aneurysms; * Men and women and persons of diverse ethnicities; * Classifications that have been confirmed using standardized methods of diagnosis with MRI verification of type and size; * Aneurysms must fall within the targeted size range.

EXCLUSION:

* Persons under the minimum age qualification to limit familial disease factors; * The presence of any thrombopholia disorders; * Indicators of controls with previous CNS bleeding; * Persons with cerebral aneurysms diagnosed in ways outside of the allowable parameters; * Unconscious or intimated patients.

Statins are available in many forms (brands) and strengths. Since there is some uncertainty of the anti-inflammatory effect among different groups for different types, we will restrict the group primarily to those who were using simvastatin, and we will collect the cases and determine whether they are at 20 or 40 mg daily dose. The original researchers noted the importance of this brand and its connection to various impacts. Part of the unique nature of our study will be well coordinated communication and monitoring arrangements. These steps will ensure the statin use status of each patient at the time of selection and across the study. This fact will be important as more MRI imaging is taken during the period of statin administration.

Both the case and control group will have their initial radiology image and total cholesterol and the method of aneurysmal detection documented. They will use assessments standardized on the MRI image CT of the brain with no contrast confirmation of the bleeding among the case subjects. An adequate radiological image from the initial diagnosis must be available to measure the aneurysmal size and progression. The lipid profile also will have had to have been obtained from each subject as a baseline at the time of diagnosis, and then will be reassessed at six months and one year periods. All subjects will be classified by their cholesterol ratings within the normal to abnormal scale.

Confounding, effect modification and analysis

As with the originating study, there will still be potential confounding and effect modifying factors. We expect to use a similar approach used by the initial investigators with regard to impacts associated with age, smoking status, alcohol and drug usage, the presence of hypertension or other atherosclerotic risk factors, and known intervention medications, if appropriate. Research subjects with thrombopholia or other instances of active or past incidence of bleeding or thrombosis with be excluded because of the known linkage of this condition to potential ruptures, often resulting in early death.6 This was the same logic used in the initial study and seems sound at this point.

Since the gold standard for looking at confounding factors centers on the calculation of Odds Ratios, we have anticipated the use of this mechanism. Our efforts will very similarly reflect what Yoshimura, Murakami, et al., did. A replication of these efforts along with the use of other multivariable regression analyses will provide us good parameters for excluding confounding situations. More formal logistical regressions or Breslow Day calculations may permit us to determine whether our results match those found previously and/or what factors across the data collected at different sites may be compared. Reviews of these results as well as the availability of chi square and data softeners should provide us a valid foundation for understanding our results.

We anticipate, as the original researchers did, that certain considerations with have important effect implications. To anticipate and measure these results, we will be categorizing subjects in regards to factors with known indicators. Subjects who regularly and routinely utilize aspirin regimens, for example, can be reviewed in comparison to those with limited or no regular use. Findings based on these reviews can be assessed linearly against known gender and age variations to help account for positive and negative multipliers.

Because we are looking into the relationship between statins and the eventual rupture of cerebral aneurysms, we have not extended the reach of our study to other levels of assessments. But it may be possible that at least two (or one broad) socio-biological factor may be relevant but that we are not looking at. This is a limitation tied to the condition but one that we believe can be tested later using our techniques.

SAH’s have direct connections to weight and exercise activities, and particularly for people who are severely overweight or obese.7 Some studies seem to suggest that statins are mistakenly seen as an opportunity to eat more poorly since cholesterol concerns are supposedly addressed by the medications. At the same time, others take their statin medications irregularly because of challenges caused by the medication. Muscular pain, for example, has been tied to popular statins such as Simvastatin, which we will be looking at directly. Whether or to what extent such social or biological/physiological facts can play in the likelihood or aneurysm ruptures is not addressed by our study. Gaining or losing weight; stopping or irregularly exercising; on-again, off-again impacts on the body could have mitigating or exasperating influences on what statins do or don't do. We have opted to not address these issues here.

One of the most important aspects of the original study centers on the fact that few human studies have been conducted to explore the interactive relationship between statins and aneurysm ruptures. Once these direct relationships or causal links are understood, it may be possible to carry the finds forward into socio-bio domains like those affecting statins and weight, exercise, obesity, etc. The systems that we expect to put in place to deal with gathering and judging data collected across hospitals (and hospital record keeping) could ensure that medical and health or wellness factors are shared too. This will open new avenues for future exploration. Not looking at these factors now does not weaken our model.

Bias and chance effects

The modifications being proposed in this study were specifically designed at address some of the biases and chance effects reflected in the original work. Most of those come from case-control models and from the limitations associated with carrying studies across locations, times and data collection systems.

This being said, even my proposal is subject to some problems relating to recall bias. The recruitment of case and control subjects still takes place in multiple settings with medical and health systems and practices that are not and cannot be fully compatible. Much information is still dependent upon the recall of patients and clinical staff or subject to interpretation as quantified for the study model. Missing or inconsistent family or illness information could also introduce concerns. The random assignment of patients for the control population could help with this.

I have factored into the design ways to compensate for this. Double checking and double reviews of files will be part of the regimen we establish, and these expectations will be reinforced in the regular communications we undertake. The use of dosage measurements and measurements of the size of the aneurysm will serve as counterbalancing these biases. In the same way, by establishing and making known clearer determinations of who can and who cannot be included in the study will enable us to minimize the influence of systemic biases or chance effects.

As referenced above, I anticipate using the study process to look at ways to align with the electronic patient data sharing activities that are already happening. By conducting the study in the US following the initiation of the Affordable Care Act, these very factors are being considered already and the study will benefit. The consent requirements will be standardized across the hospital locations while conforming to HIPPA and other privacy considerations.

Remaining study limitations

The investigators in the original study identified four limiting elements of their work:

* The inherent problems associated with case-control studies; * Their decision not to analyze serum lipids, cholesterol and like biomarkers tied directly to statin use; * The still imperfect result of using miltivariate logistic regression and other balancing methods to compensate for confounding influences; and, * The inefficiency of the record-keeping, recall and statin use patterns of control patients.

The steps proposed in this adaptation deal with most of these directly. I continue to utilize the case-control model because it remains a recognized approach for studying a rather uncommon and infrequent condition. This is particularly true when paired with more sophisticated data adjusting or softening capabilities that can compensate for information collected across time periods or even different data sets.

More importantly, I believe I have advanced the benefits of the study by directly addressing factors tied to record compatibility, sharing of data, and statin usage and dosage considerations. By establishing baselines for serum lipids and cholesterol levels, we will have opened changes in these indicators to better review and alignment with possible causal links. In a similar way, by requiring the identification of the size of aneurysm as a study determinant, we will be able to document measurable progress on protecting the site from rupture.

Substantially increasing the size of the subject populations (both case and control) can help compensate for the study problems of non-random investigations. We do this at the risk of adding other site specific biases or effect modifiers. While we can control for some of them with statistical adjustments, others demand more consistent data collect. The development of the clear standards and of the regular communications and monitoring elements will likewise ensure that data capturing is sound and readily usable.

Scientific impact

Yoshimura, Murakami, et al. concluded that their findings provided one of the first levels of empirical validation of the association between statin usage and not having cerebral aneurysms advance toward rupture. They did so using a reputable study model. But even so, their efforts contained some notable limitations. These qualifiers don't take away from their results but do suggest ways that their work can be improved.

The expansions and refinements suggested here demonstrate how this can happen. They include precise systemic adaptations that can make it easier to collect reliable data, even across a larger set of data gathering locations. They also show how this information can be tied to precise biomarkers that may indicate why they can reduce the possibility of ruptures. This will be very important in developing clinical interventions other than surgery.

Just as critical, however, is the fact that the study grows and is refined by way of using advancing technologies like electronic patient file sharing. This is a benefit that may extend far beyond this study itself as technology also gives us a deeper look into what medications can do.

Section C: Power and Sample Size (extra credit)

I think at a minimum you can run one of the formulas and then explain the values set. Hard for me to do that without making assumptions I don’t know.

Section D: References

1. Hae Woong Jeong,MD,Sang-il Suh,MD. Clinical Practice Guideline for the Management of Intracranial Aneurysm. Neurointervention 2014;9:63-71. 2. Rinkel GJ. Journal of Neuroradiology. Natural history, epidemiology and screening of unruptured intracranial aneurysms. 2007.11.004 3. White PM,Wardlaw JM. Unruptured intracranial aneurysms. J Neuroradiology.2003;5:336-50. 4. Nicholas C Bambakidis,MD and Wareen R.Selman. Statins and aneurysm. J Neurosurg.2012;116:626-637. 5. NIH. Accessed 12/2/2014: http://www.ninds.nih.gov/disorders/cerebral_aneurysm/cerebral_aneurysms.htm 6. Theodora W.M Raaymakers, Gabriel J.E Rinkel, Ale Algra. Mortality and Morbidiity of surgery of Unruptured intracranial Aneurysm. A meta –Analysis.Stroke. Am H Association journal.1998;29:1531-1538. 7. Rita F. Redberg, MD, MSc,JAMA Internal Med 2014; 174(7):1046.