Prevention of Postoperative Cognitive Dysfunction by Minocycline in Elderly Patients after Total Knee Arthroplasty: A Randomized, Double-blind, Placebo-controlled Clinical Trial

This double-blind, prospective, randomized controlled trial, designed to evaluate the effectiveness of minocycline in preventing postoperative cognitive dysfunction, was conducted at two centers in Japan. The study conformed to the standards of the Declaration of Helsinki and was approved by the ethics committees of both facilities (Gunma University Hospital Institutional Review Board registry identifier 1070; Gunma Central Hospital Institutional Review Board registry identifier 2014-001; Maebashi, Gunma, Japan) and was registered with the University Hospital Medical Information Network Clinical Trials Registry (UMIN000014620) on July 24, 2014. Written informed consent for study participation was obtained from all patients and nonsurgical controls before enrollment. The primary endpoints were the incidences of postoperative cognitive dysfunction at 1 week and 3 months after surgery. No statistical adjustments were made for interpreting multiple primary outcomes.

The study included patients between the ages of 60 and 90 years undergoing total knee arthroplasty under general anesthesia between January 2015 and October 2018. Patients who met the following criteria were excluded from the study: (1) American Society of Anesthesiologists (ASA) Physical Status III or higher; (2) a history of mental disease, drug dependency, or alcoholism; (3) liver or renal disorders that can affect the pharmacokinetics of minocycline; (4) unable to discontinue medication with possible interactive effects with minocycline, including digoxin, methotrexate, warfarin sodium, and sulfonylurea-based hypoglycemic drugs; (5) patients with dementia before surgery; and (6) physician’s judgment that participation in the study was inappropriate. A previous study that examined the incidence of postoperative cognitive dysfunction in elderly patients after total knee arthroplasty and total hip arthroplasty found that the incidence of postoperative cognitive dysfunction 1 week after surgery was 9.1%.¹⁸  A previous preclinical study showed that prophylactic administration of minocycline almost completely (greater than 90%) suppressed the impairment of recall of fear memories after surgery.¹¹  Thus, we estimated that the incidence of postoperative cognitive dysfunction in the minocycline group would be reduced to 0.6% (approximately 93% reduction). Based on this estimate, with an α of 0.8 and a β of 0.2, the total number of cases required in this study was 180; allowing for a 10% dropout, 200 patients were recruited.

A total of 30 nonsurgical controls aged greater than 60 yr old, whose age and sex were similar to those of the patients, were subjected to the same tests between December 2014 and March 2019 to examine the learning effect of repeated testing.¹⁹  The exclusion criteria for controls were as follows: (1) ASA Physical Status III or higher, (2) subjects with dementia, (3) subjects who had undergone surgery within 6 months of the study or planned to undergo surgery during the study period; (4) subjects with a history of neurologic or psychiatric disorders, and drug or alcohol dependence, and (5) subjects deemed unfit for study participation based on the physician’s judgment. The recruitment of controls was outsourced to YMG Support Co., Ltd. (Japan). A comparison of the background characteristics of patients and controls is shown in supplemental table S1 (https://links.lww.com/ALN/C969).

After enrollment, eligible patients were randomly assigned 1:1 to either the minocycline group or the placebo group. Randomization was done using a central randomization system managed by the Internet Data and Information Center for Medical Research. Before randomization, the patients were first stratified for the following five factors that could have biased the results: (1) age; (2) sex; (3) possibility of pre-existing mild cognitive impairment, based on Montreal Cognitive Assessment scores of 25 points or less; (4) previous history of total knee arthroplasty; and (5) hospital. Stratification was performed using a central randomization system using prepopulated patient information. Nichi-Iko Pharmaceutical Co., Ltd. (Japan) provided the capsule formulation of minocycline and placebo. Patients assigned to the minocycline group took minocycline capsules (100 mg) in the evening on the day before surgery, the morning of surgery, and in the morning and evening on postoperative days 1 to 7. Patients assigned to the placebo group took a placebo designed to look identical to minocycline in terms of size, shape, and color on the same schedule. The minocycline or placebo tablet was personally handed to each patient by the ward nurse. After the patient had taken the medication, the press-through pack was collected and was checked for omissions by the pharmacist. This approach ensured adherence to the study regimen. To ensure that the drugs assigned to each group were correct, they were sampled and analyzed for their components by a mass spectrometer (Xevo TQ-MS, Waters Corporation, USA). The results of the analysis are shown in supplemental figure S1 (https://links.lww.com/ALN/C970). Both the patients and endpoint assessors were blinded to the allocation.

All subjects received general anesthesia. Anesthesia was induced with propofol, remifentanil, and rocuronium and was maintained with desflurane (4 to 7%) and remifentanil to maintain the Bispectral Index between 30 and 60. The patients were mechanically ventilated after intubation. The ventilator settings were adjusted to keep end-tidal carbon dioxide at 30 to 40 mmHg. Fraction of inspired oxygen (Fio₂) was usually maintained between 0.4 and 0.6. Hypoxia was defined as percutaneous oxygen saturation less than 90%, and the Fio₂ was increased as needed if hypoxia occurred. During anesthesia, arterial blood pressure was monitored every 2.5 min. Ephedrine (4 to 8 mg) or phenylephrine (0.1 mg) was given to maintain mean arterial pressure at 60 mmHg or higher. Atropine 0.5 mg was given to maintain the pulse rate at 45 beats/min or higher. At the end of the surgery, sugammadex (2 mg/kg) was administered before extubation. All patients received standardized postoperative care and rehabilitation.

We performed Montreal Cognitive Assessment preoperatively on all patients who agreed to participate in the study. The purpose of Montreal Cognitive Assessment was to exclude patients with pre-existing dementia and determine the presence of preoperative mild cognitive impairment.²⁰  The Japanese version of the Montreal Cognitive Assessment has been previously reported as being able to diagnose mild cognitive impairment with excellent accuracy, with a sensitivity of 93% and specificity of 87% when the cutoff value is set at 25, the same as that of the original Montreal Cognitive Assessment.²¹  Considering the lowest Montreal Cognitive Assessment score for patients with mild cognitive impairment,²⁰  patients with a Montreal Cognitive Assessment score of less than 15 were diagnosed with dementia and excluded from the study.

A test battery consisting of four cognitive function tests was performed three times: at the time of patient enrollment and at 1 week and approximately 3 months after surgery (range, 1 to 5 months postoperatively). Postoperative cognitive dysfunction was diagnosed in the same manner as in the previous study.²²  The test battery has been shown to work well with postoperative patients and is feasible to perform in a clinical setting.^2,14,18  The first test was the Visual Verbal Learning Test, based on Rey’s auditive recall of words. A list of 15 words had to be learned after their presentation at a fixed rate (2 s/word) on drawing paper. The patients were asked to recall as many words as possible. The Visual Verbal Learning Test was repeated three times with different sets of words. A fourth Visual Verbal Learning Test was conducted at the end of the neurocognitive test battery. The second test was the Trail Making Test Part B. Numbers and letters were written on the inspection sheet, and the subject was asked to connect the numbers and letters alternately with lines using a ballpoint pen. The Trail Making Test indicates the subject’s visuomotor skills and short-term memory. The third test was the Stroop Color and Word Test to evaluate selective attention and interference susceptibility. In this test, the subject was shown a word that had a mismatched combination of word and ink color and was asked to choose the word corresponding to the color of the ink in which the word was written. The fourth test was the Letter-Digit Coding Task to evaluate information processing speed and accuracy. Patients were shown the relationships between the numbers from 1 to 9 and corresponding symbols and were asked to write the corresponding symbols below the numbers printed on the question paper. Additionally, a Mini-Mental State Examination containing various elements was performed before the test battery to evaluate the effectiveness of a composite but simple neurocognitive test. Further, the presence or absence of depressive symptoms was evaluated on the geriatric depression scale before the test battery, since depression can affect cognitive test results.²² 

All the tests were carried out in a quiet room by trained workers blinded to the patient’s group allocation. The same subject was examined by the same examiner three times for all the tests. Tests were usually performed during the daytime (9 am to 3 pm). Cognitive function was evaluated based on eight variables from the four tests: averaged number of words recalled during three trials of the Visual Verbal Learning Test, the number of words recalled with delayed recall in the Visual Verbal Learning Test, and the time and number of errors in the Trail Making Test, Stroop Color and Word Test, and Letter-Digit Coding Task.^2,14,18 

For all patients, we also investigated the confusion assessment method for the intensive care unit (ICU), which is used to assess delirium in patients in the ICU, three times a day, from the night of surgery to the night of the third day after surgery.²³  We also recorded a numeric rating scale to evaluate postoperative pain. Numeric rating scale was examined three times a day from the night of surgery to the night of the seventh day after surgery. Both confusion assessment method for the ICU and numeric rating scale were evaluated by pretrained ward nurses based on patient interviews. Additionally, we estimated the activity of daily living score 3 months after surgery using five questions related to shopping, domestic work, preparation of meals, bodily care, and dressing.¹⁴  The examiner who conducted the neurocognitive tests conducted the interviews for this estimation.

The presence or absence of adverse events was assessed from the start of the oral administration of the study drug to the seventh postoperative day. Adverse events were graded by Common Terminology Criteria for Adverse Events version 4.0. Serious adverse events were reported to the hospital director and clinical trial review board.

Fisher’s exact test was used analysis of the primary outcomes and nominal variables and binary variables in the patients’ backgrounds. The Wald test was used for multiple imputation and logistic regression analysis. The significance level was set at 5% or less for all tests.

The Z scores of the patients’ neurocognitive test results were calculated.²²  First, we compared changes in performance among controls for each test from baseline (the first session) to one week and three months after the first session and calculated the means and SD of these differences. The mean changes were considered to represent learning effects. We then compared baseline scores with 1-week and 3-month test results in the patients, subtracted the average learning effect from these changes, and divided the result by the control-group SD to obtain a Z score for each test. Large positive Z scores indicate deterioration in cognitive function from baseline in patients compared with controls. The composite Z score was calculated as the sum of Z scores divided by the SD for this sum of Z scores in controls. Patients with a composite Z score > 1.96, or more than two tests with a Z score > 1.96, were diagnosed as having postoperative cognitive dysfunction. This calculation method is the same as that used in many previous studies, and its validity has been confirmed.^14,18,24 

The multiple imputation method and the worst-rank method²⁵  were used to estimate missing values for the third neurocognitive tests. The variables used for multiple imputation included the variables used for stratified randomization plus educational history, first Mini-Mental State Examination, and timing of the third neurocognitive tests. Multiple imputation was repeated 100 times, and the results were integrated and compared between groups by the Wald test.

Statistical data processing was outsourced to WDB Clinical Research Co., Ltd. (Japan). Statistical analysis was performed using SAS 9.4 (TS1M7 DBCS3170) for Windows (64 bits) and confirmed by authors who are experts in statistics.

A total of 576 patients underwent total knee arthroplasty during the study period (fig. 1). Of these patients, 374 were excluded due to failure to meet inclusion criteria or declining participation, and 202 patients were randomly assigned to the minocycline or placebo group. Of these 202 patients, 10 patients in the minocycline group and 7 in the placebo group were excluded from the study because they refused to either take the test drug or undergo the second neurocognitive tests. The study ultimately included 185 patients with a median age of 75 yr. The characteristics and intraoperative data of the patients in the two groups were similar (table 1).

A total of 12 patients were diagnosed with postoperative cognitive dysfunction 1 week after surgery (12 of 185, 6.5%) by the neurocognitive tests. There was no significant difference in the incidence of postoperative cognitive dysfunction between the minocycline group (8 of 90 [8.9%]) and the placebo group (4 of 95 [4.2%]) at 1 week after surgery (odds ratio, 2.22 [95% CI, 0.64 to 7.65]; P = 0.240; table 2). Since 20 patients could not be tested 3 months after surgery (fig. 1), the incidence of postoperative cognitive dysfunction was calculated by the following three analysis methods (table 2): (1) The presence or absence of postoperative cognitive dysfunction was estimated from the patient’s background and the results of preoperative cognitive function tests. The third examination was assumed to have been performed on postoperative day 85 to 94 (multiple imputation). (2) Patients who could not be tested were not included in the analysis (no imputation). (3) Patients who could not be tested were judged to have postoperative cognitive dysfunction based on the worst-rank score method.²⁵  The incidence of postoperative cognitive dysfunction 3 months after surgery calculated by the multiple imputation method showed no difference between the minocycline group (15.3 of 90 [17.0%]) and the placebo group (15.3 of 95 [16.1%]; odds ratio, 1.07 [95% CI, 0.49 to 2.32]; P = 0.889). In contrast, when no imputation was adopted, the incidence of postoperative cognitive dysfunction after three months was 14 of 165 (8.5%), with a higher incidence of postoperative cognitive dysfunction in the minocycline group (11 of 82 [13.4%]) compared to the placebo group (3 of 83 [3.6%]; odds ratio 4.13 [95% CI, 1.11 to 15.40]; P = 0.027). By the worst-rank method, 34 of the 185 patients (18.4%) were diagnosed with postoperative cognitive dysfunction 3 months after surgery, with no difference in the incidence of postoperative cognitive dysfunction between the minocycline (19 of 90 [21.1%]) and placebo groups (15 of 95 [15.8%]; odds ratio, 1.43 [95% CI, 0.68 to 3.02]; P = 0.448). Supplemental tables S2 and S3 (https://links.lww.com/ALN/C969) show the results of the neurocognitive test battery in the patients and control subjects.

The median interval from surgery to the second neurocognitive tests in the entire patient cohort was 7 days and to the third neurocognitive tests was 91 days (interquartile range, 81 to 95 days). There were no significant differences between the two groups in the number of days from surgery to the second and third neurocognitive tests (supplemental table S4, https://links.lww.com/ALN/C969).

Univariate logistic regression analysis revealed the following 6 factors as risk factors for postoperative cognitive dysfunction 1 week after surgery: patient age, the hospital where surgery was performed, preoperative Montreal Cognitive Assessment results, total knee arthroplasty history, educational level, and preoperative Mini-Mental State Examination results (table 3). Three months after surgery, minocycline and a low preoperative Montreal Cognitive Assessment score were risk factors with no imputation, while a low preoperative Montreal Cognitive Assessment score was a risk factor for postoperative cognitive dysfunction when the worst-rank score method was adopted (table 4).

Table 5 shows the results of analysis of the secondary outcomes. Only eight patients developed delirium within 3 days after surgery, with an incidence of 4.3% (8 of 185). There were no significant differences in the incidence of postoperative delirium between the two groups. Further, there were no differences between the two groups in the maximum and mean numeric rating scale of postoperative pain examined for 7 days postoperatively. The type and total amount of drugs used for postoperative analgesia did not differ between the two groups, as shown in supplemental table S5 (https://links.lww.com/ALN/C969). The median total activities of daily living score was 3.0 in both groups, indicating that many patients had improved activities of daily living after total knee arthroplasty.

Thirteen patients discontinued medication at their own request. Four other patients dropped out during the observation period. Two of these cases had subarachnoid hemorrhage and acute exacerbation of interstitial pneumonia, respectively, requiring intensive treatment. These cases were reported to the hospital director and the clinical trial review board. The remaining two patients dropped out on postoperative days 1 and 2 due to elevated liver enzymes and dizziness/vomiting, respectively. Both patients had been assigned to the minocycline group. Supplemental table S6 (https://links.lww.com/ALN/C969) shows the results of adverse event monitoring. The patients included in the table were able to complete at least 7 days of medication, because the adverse events, if any, were minor.

This study suggests that minocycline is ineffective in preventing postoperative cognitive dysfunction. Univariate logistic analysis indicated the association between low preoperative Montreal Cognitive Assessment score and the development of postoperative cognitive dysfunction 3 months after surgery.

Despite promising animal studies, it is unclear why this study failed to show a preventive effect of minocycline on postoperative cognitive dysfunction. However, it is not uncommon for research results on laboratory animals and humans to differ.^26,27  The other possible reasons for the differences are discussed below. This study was designed so that previously reported risk factors for postoperative cognitive dysfunction were similar in the minocycline and placebo groups. Therefore, it is very unlikely that there was a bias in the potential postoperative cognitive dysfunction risk between the two groups. Further, it is unlikely that there were errors in study drug administration relative to group assignment. From the results of random sampling of the test drugs after allocation, it is implausible that the test drugs in both groups were accidentally replaced. The dose of minocycline in this study, 200 mg/day, might be debatable. Since this dose is recommended in the package insert, administering more than this was untenable from a patient safety perspective. Another possibility is that the administration period of minocycline might have been too short. Indeed, a recent preclinical study suggested that minocycline suppresses microglial activation immediately after surgery but has no long-term effects and might not ultimately suppress postoperative cognitive dysfunction.^7,28  We assumed that intraoperative cytokine- induced inflammation in the brain would subside within 1 week after surgery, and therefore, minocycline was administered for only 1 week. Since there are no previous studies, including preclinical studies, determining how long postoperative neuroinflammation should be suppressed to reduce postoperative cognitive dysfunction, further research on this topic is needed. A database search indicated that a study with the same objective is currently underway (NCT02928692). The results of that research will also be noted.

Missing data is inevitable in studies with many patients. In this study, 20 patients did not undergo the third examination. When missing data are considered “informatively missing,” a worst-rank score method²⁵  and reference-based multiple imputation are recommended for imputation of missing data and has been used in previous studies.^29–31  In the worst-rank method, patients with missing data were treated as having developed postoperative cognitive dysfunction. With this approach, there was no difference in the incidence of postoperative cognitive dysfunction after 3 months between the two groups. We also estimated the missing values by multiple imputation and found no difference in the incidence of postoperative cognitive dysfunction between the two groups. However, when the method of no imputation was adopted, the minocycline group had an even higher incidence of postoperative cognitive dysfunction. Considering the results derived from these three analysis methods, it is highly likely that minocycline has, at best, no preventive effect on postoperative cognitive dysfunction.

Many studies have investigated the incidence of postoperative cognitive dysfunction, although the results vary from study to study.^14,17,18,32  This variation might be due to differences in study settings, including patient background characteristics, surgical procedures performed, and postoperative cognitive dysfunction assessment methods. The current study was unique in that the average age of the patients was 75 yr, which is much older than in previous studies on postoperative cognitive dysfunction.³²  It should be noted that the incidence of postoperative cognitive dysfunction at 1 week was 6.5%, which was lower than that in previous studies,^14,17,18  even though old age is considered a risk factor for postoperative cognitive dysfunction. Further, it is important to note that since the percentage of patients diagnosed with mild cognitive impairment by preoperative Montreal Cognitive Assessment was as high as approximately 80%, our findings might be most applicable to a somewhat cognitively impaired population. Other studies conducted in Japan, in which the average age of subjects was about 75 yr, showed similar Montreal Cognitive Assessment scores, suggesting that the result reflects old age.^33,34  In this study, since we adopted a diagnostic method for postoperative cognitive dysfunction that has been demonstrated to have excellent sensitivity and specificity in previous studies,^2,14,18  we believe that the relatively low incidence of postoperative cognitive dysfunction after 1 week despite the high percentage (greater than 80%) of patients with preoperative cognitive impairment was not due to a sensitivity issue. Additionally, the incidence of postoperative cognitive dysfunction at 3 months was higher, ranging between 8.5 and 18.4%, depending on the handling of missing data. Given that the incidence of postoperative cognitive dysfunction might be overestimated when the worst-rank score method is adopted, the incidence of postoperative cognitive dysfunction after 3 months was comparable to that in past studies, because the pooled incidence of postoperative cognitive dysfunction after 3 months is reportedly 11.7%.³²  In 2018, after we designed the study, the Perioperative Cognition Nomenclature Working Group announced a new nomenclature.³⁵  Applying that nomenclature here, postoperative cognitive dysfunction after 1 week and 3 months in this study could be defined as “delayed neurocognitive recovery” and “postoperative neurocognitive disorder,” respectively.

This study has the following limitations. First, all the subjects were patients who had undergone total knee arthroplasty. We limited the patients to a single surgical procedure to eliminate bias due to different surgical procedures. Further, we excluded ASA Physical Status III and higher patients, which, while ensuring patient safety, limited the generalizability of the study results. Hence, it is unclear whether the results of this study are applicable to all postoperative patients. Second, the control group included elderly people who did not need knee surgery. Patients undergoing total knee arthroplasty were likely to have restricted daily activities due to long-term knee pain. Since the effect of knee pain itself or that of the resultant activity restriction on the function of the upper central nervous system is unknown, the learning effect in patients in this study is not guaranteed to be equivalent to that in the control elderly subjects. In other words, the incidence of postoperative cognitive dysfunction might have been overestimated because of an overestimated learning effect in this study. However, even if this were the case, it would not have affected the efficacy comparison of the study drugs. Third, although the power analysis showed that a total of 180 cases were required, only 165 patients underwent cognitive function tests 3 months after surgery, which is underpowered. To solve this problem, we needed to estimate missing data, which we initially had not planned to do. We found that studies involving older adults, especially those with pre-existing cognitive decline, need to overestimate the number of dropouts. Fourth, we performed stratified randomization using five known risk factors for postoperative cognitive dysfunction to eliminate differences in patient backgrounds between the two groups. However, the number of risk factors for stratification might have been too large compared to the sample size of this study, and it would have been preferable to use minimization methods. However, this might not pose a significant problem as the patient backgrounds in both groups were similar, as shown in table 1.

In conclusion, we showed that minocycline did not have a protective effect on postoperative cognitive dysfunction in this study. Additionally, Montreal Cognitive Assessment might be useful as a simple preoperative diagnostic tool for identifying patients at risk of developing postoperative cognitive dysfunction.

The authors thank Nagisa Nakano, B.Sc. (Department of Anesthesiology, Gunma University Graduate School of Medicine, Maebashi, Japan), Sachiko Kurose, B.Sc., Naoko Fugami, B.Sc., Ayumi Kobayashi, B.Sc., and Etsuko Sanada, B.Sc. (Department of Neurology, Gunma University Hospital) for conducting neurocognitive tests. Takashi Osawa, M.D., Ph.D., Hiroyuki Shiozawa, M.D., Ph.D., Takuya Omodaka, M.D., Ph.D., and Shogo Hashimoto, M.D., Ph.D. (Department of Orthopedic Surgery, Gunma University Hospital, Maebashi, Japan) for patient recruitment. Kumi Katagi, B.Sc., Tomomi Nomura, B.Sc., Sahoko Iino, B.Sc., and Miho Aoki, B.Sc. (Department of Anesthesiology, Gunma University Graduate School of Medicine) for patient registration. Mie Araki, B.Sc., Noriko Ito, M.H.Sc., Sachiko Chigira, M.S., and Shoichi Yamaga, B.Sc. (Department of Pharmacy, Japan Community Healthcare Organization, Gunma Chuo Hospital, Maebashi, Japan) for test drug preparation. Jiro Kamiyama, M.D., Ph.D. (Intensive Care Unit, Gunma University Hospital) and Shinya Sakamoto, M.D. (Department of Anesthesiology, Gunma University Hospital) for explaining the research to the patients. The former affiliation is displayed for those who moved in the middle of the study.

This study was funded by the Japan Society for the Promotion of Science KAKENHI (grant No. 15K10533) and Daiwa Securities Health Foundation (Tokyo, Japan).

The authors declare no competing interests.

Full protocol available at: takazawt@gunma-u.ac.jp. Raw data available at: takazawt@gunma-u.ac.jp.

Supplemental Tables, https://links.lww.com/ALN/C969

Supplemental Figure, https://links.lww.com/ALN/C970