Relationship Between Smoking and Pressure Injury Risk: A Systematic Review and Meta-Analysis

Pressure injury (PI) is an adverse event in hospitalized patients.^1-4 It causes discomfort and pain, impairs the patient’s quality of life, prolongs length of hospital stay, increases costs, and has been associated with high morbidity and mortality rates, thus creating negative psychological and physical effects on patients and their families.^5-9 The development of PI depends on multiple extrinsic and intrinsic risk factors^10,11 including age, exercise, diabetes, and alcohol or drug use.^12-16 It has been shown that the implementation of PI prevention guidelines or care bundles can help to prevent PIs by targeting intrinsic and extrinsic risk factors associated with PI development. An interventional study showed that PI prevalence in a hospital in Denmark was reduced by 50% after implementing care bundles for 6 years.¹⁷ Another interventional study showed that a bundle of 7 interventions lowered hospital-acquired PI rates and that the rates continued to remain below 1%.¹⁸ Therefore, it is necessary to understand and identify PI risk factors to provide preventive interventions and efficiently allocate resources in patient care.

It is acknowledged that smoking is a risk factor for vascular diseases.^19,20 However, it is questionable whether smoking is an extrinsic risk factor for PI. In a prospective study that analyzed 350 participants with spinal cord injury, it was found that those who smoked (n = 36) had a greater risk of PI (odds ratio [OR] = 2.69; 95% confidence interval [CI], 1.00–7.27).²¹ A retrospective cross-sectional study by Lin et al²² of 225 patients in the intensive care unit of a hospital in China found former smoking to be a risk factor of PI (OR = 3.054; specificity = 85.8%). Rabadi and Vincent²³ performed a retrospective review of the electronic medical records of veterans and found that 11 of 37 current smokers had a PI and that 16 of 50 former smokers had a PI; the incidence of PI in these 2 groups was 29.7% and 32%, respectively. The authors concluded that there were no significant differences between the PI group and non-PI group relative to smoking status.

The contradictions of the above-mentioned studies could be due to small sample sizes, residual confounding from population heterogeneity, and regional differences. Hence, a systematic review and meta-analysis of international studies is needed. The purpose of this study was to examine the relationship between smoking and PI risk.

Database search. The authors conducted a comprehensive search of PubMed, Web of Science, and China National Knowledge Infrastructure (CNKI). The search terms included the key words “pressure injury,” “pressure injuries,” “pressure ulcer,” “decubitus ulcer,” “smoke,” “cigarettes,” “tobacco,” and their possible combinations. Searches were implemented on abstracts, key words, and free text words according to the database. Time of publication was 2001 through July 20, 2020.

Inclusion and exclusion criteria. Inclusion criteria for the studies were as follows: 1) patients: community or hospital-based adult patients, 2) study: cohort study or case–control study including PI patient group and non-PI patient group, 3) comparison: patients who smoke currently or in the past and patients who do not smoke, 4) study design: retrospective or prospective, 5) participant characteristics: age and sex and 6) study quality: moderate level and above on the Newcastle-Ottawa Scale (NOS).

Exclusion criteria included the following: 1) no complete raw data, 2) reported results included PI and other wounds for which the exact incidence of PI was not known, and 3) case reports and reviews. For the research from the same institution and author, only the latest reports were included. Two (2) reviewers assessed a study independently by screening the title, abstract, and full text. When a result was disputed, a third investigator was responsible for coordination.

Data extraction. For each eligible study, the authors extracted the following information: 1) name of the first author, 2) country, 3) years of patient data included and publication, 4) study design, 5) sample size, 6) mean age, 7) sex, 8) factors related to the primary wound (the underlying reasons for PI development), 9) PI site, 10) PI stage, 11) diabetes mellitus (DM) prevalence, and 12) the number of smokers and nonsmokers among patients with or without PIs. The above data were first extracted independently by one author, and then the accuracy of all the extracted data was checked by another author.

Study quality assessment. The NOS,²⁴ which is a validated tool to evaluate risk of bias in studies, was used by 2 reviewers independently to assess the quality of included cohort studies. The 3 major domains on the NOS are selection (0–4 points), comparability (0–2 points), and exposure (0–3 points). Based on a score ranging from 0 to 9, the quality of the studies was classified as poor (score of 0–3), moderate (score of 4–6), or high (score of 7–9).

Statistical analysis. Statistical analyses were accomplished by using Stata software (StataCorp). ORs and corresponding 95% CIs were used to quantify the relationship between the incidence of PI and smoking with an expression of a dichotomous variable.²⁵ Heterogeneity among studies was represented by Higgins I-squared (I²).²⁶ The fixed-effect model was applied when heterogeneity was not substantial (I² ≤ 30% or P > .05); otherwise, the random effect model was used to provide more conservative computations (I² > 30% or P ≤ .05).²⁷

Visual symmetrical distribution of the funnel plot, jointly Begg’s adjusted rank correlation test and Egger’s regression test, were used to estimate potential publication bias.^28,29 Subgroup analyses were performed to examine the underlying covariates that contributed to heterogeneity, which included PI site, PI stage, and study design.²⁹ Sensitivity analyses were performed to assess the influence of each study on the entire effect.³⁰

The authors of some of the above studies had performed logistic regression to control for the clinical risk factor of smoking, and their adjusted ORs and 95% CIs were used for statistical analysis. Meta-regression was conducted to explore the causes of heterogeneity. Specifically, the independent variables that may be related to heterogeneity, including year, incidence of PI, mean age of patients, and DM prevalence, were explored.

Search results. A total of 1552 records were identified through database searches, of which 346 duplicates were excluded. When screening titles and abstracts, 882 studies were excluded due to an irrelevant topic. Among the remaining 324 full-text papers, 31 met the inclusion criteria and were assessed for eligibility, although 16 of them provided insufficient information and were subsequently excluded. Thus, 15 studies were eligible for meta-analysis (Figure 1).

Study characteristics. The eligible studies (3 prospective and 12 retrospective; Table 1, Part 1 and Part 2) were conducted between 2008 and 2020 in the United States,^21,31 Norway,¹⁴ China,^13,22,32-39 Japan,⁴⁰ and Singapore.⁴¹ Patient data were collected between 2007 and 2019. A total of 11 304 patients were included in the current study (5721 males and 5583 females, age range: 23–100 years old; 910 patients with PI and 10 394 without). The sample sizes for the studies ranged from 60 to 6912. All included studies were of case–control design. The underlying reasons for PI development were related to surgery,^32-38,41 medical devices,^39,40 spinal cord injury,^14,21,31 and stay in an intensive care unit.²² Pressure injury location, PI stage, and DM prevalence were reported for all included studies.

Quality assessment. Of the 15 eligible studies, 13 were considered to be of higher methodological quality (NOS scores ranging from 7 to 9).^14,22,31-41 The remaining 2 studies were considered to be of moderate quality (NOS score, 6^13,21) due to an insufficient characterization of samples or lack of adjustment for confounding variables. The median overall NOS score was 8.0, which demonstrated that the studies were of general high quality.

Meta-analysis. As shown in Figure 2, because of significant heterogeneity (I² = 64.0%; P = .00), the authors applied a random effects model for all 15 studies. The results showed that current and former smokers had a 1.5 times higher odds of developing PI than nonsmokers (OR = 1.498; 95% CI, 1.058-2.122).

As shown in the funnel plot in Figure 3, all included studies were symmetrically distributed. Begg’s test (P = .322) and Egger’s test (P = .666) were used to test publication bias. Both P values of Begg’s test and Egger’s test were greater than 0.05 and therefore showed that there was no publication bias.

Subgroup analyses for risk estimates caused by study design between smoking and PI for retrospective and prospective studies are summarized in Figure 4. The OR for the retrospective studies was 1.607 (95% CI, 1.043-2.475; I² = 68.8%; P =.00) and was statistically significant. The OR for the prospective studies was 1.18 (95% CI, 0.735-2.017; I² = 30%; P = .240) and was not considered to be statistically significant due to the small number of studies.

Sensitivity analysis was conducted by removing one study at a time, and the results are shown in Figure 5 and Table 2. Compared with the original results of all included studies, the consequence of sensitivity and 95% CI (0.983-1.827) were reduced slightly after the removal of the study by Fang,³³ and 95% CI (1.212-2.254) increased slightly after removing the study by Cai et al.¹³ Thus, it can be indicated that none of the studies strongly influenced the result and that the meta-analysis results were robust.

Pooled adjusted values. To further clarify the risk factor of smoking, the authors collected OR values and 95% CIs obtained from the studies that contained multiple logistic regression analysis. Analysis of the adjusted OR values (OR = 1.969; 95% CI, 1.406-2.757; P = .000) indicated smoking was considered an independent risk factor for the occurrence of PI (Figure 6).

Meta-regression analysis. The meta-regression analysis (Table 3) was performed to probe into possible sources of heterogeneity across the contained studies, including PI incidence, mean age of patients, years of patient data included, and DM prevalence (Figure 7) as covariates as well as predictor variables. The results of meta-regression analysis suggested differences in PI incidence (t = 0.75; P = .466; CI, -8.94-18.46), inclusion year of the cases (t = -0.48; P = .637; CI = -1.14-0.72), mean age of patients (t = -0.47; P = .650; CI, -0.41-0.27), or DM prevalence (t = 0.69; P = .509; CI = -24.56-46.00), could not account for the observed heterogeneity in this review as P values were > .05. In other words, there was no relationship between PI morbidity and PI incidence, included year, mean patient age, and DM prevalence.

In this systematic review and meta-analysis, 15 studies (11 304 patients) were analyzed to evaluate the relationship between smoking and PI. The results suggested that current and past smokers had significantly higher risk (about 1.5 times) of developing PI than nonsmokers. The pooled adjusted OR value led to the same conclusion that smoking is associated with a significant increased risk (about twice the odds) of PI. It was confirmed that this relationship was not related to study year, incidence of PI, mean patient age, or DM prevalence in each sample. These results are in line with those of Li et al²¹ and Lin et al.²²

Currently used PI assessment scales, including the Braden scale, do not take smoking into consideration as a risk factor. A systematic review and meta-analysis by Wei et al⁴²  reported that the overall weighted AUC for the Braden scale in predicting risk of PI in adult patients treated in the intensive care unit was 0.7812, which showed that the Braden scale had moderate predictive validity for PI in this patient population. In another systematic review and meta-analysis, Chen et al⁴³ showed that the overall diagnostic accuracy (Q*) of the Braden scale was 0.7090 and suggested this scale had moderate predictive validity and low predictive specificity for PIs in long-term care residents. The results of the above 2 studies indicate that it is necessary to develop and test new PI risk assessment scales for this population.

Finally, our research confirms that smoking is a risk factor for another disease, which could promote the implementation of smoke-free policies to decrease the number of smokers.^44,45

This study has several limitations. First, the majority of studies included only the number of individuals who smoked past or currently, not the number of cigarettes smoked per day. Therefore, the correlation between number of cigarettes smoked and PI cannot be explored. Second, the pooled OR was 1.498 (95% CI, 1.058-2.122) without significant heterogeneity (I² = 30%; P = 0.240) in 3 prospective studies, and the number of prospective studies was limited. Thus, further research is required to assess the associations among tobacco type, smoking frequency, and PI to confirm these findings.

This systematic review and meta-analysis showed that former and current smokers have an approximately 1.5 times higher risk of PI development than nonsmokers. Future prospective studies that have larger sample sizes and target smoking as a risk factor of PI development are needed.

Ms. Wu is an undergraduate, School of Medicine, Nantong University, Nantong, Jiangsu. Mr. Gu is a master, Yiwu Central Hospital, Yiwu, Zhejiang. Ms. Yu is an undergraduate, School of Medicine, Nantong University, Nantong, Jiangsu. Ms. Feng is a master, Nantong University Hospital, Nantong, Jiangsu. Ms. Xu is an undergraduate, School of Medicine, Nantong University, Nantong, Jiangsu. Ms. Zha is a master, Nursing Department of Nantong University Hospital, Nantong, Jiangsu. Dr. Shen is a professor, School of Medicine, Nantong University, Nantong, Jiangsu. Dr. Chen is a professor, School of Public Health, Nantong University, Nantong, Jiangsu. All institutions are in the People’s Republic of China. Address all correspondence to: Dr. Wang-Qin Shen or Dr. Hong-Lin Chen, Nantong University, Nantong, Jiangsu, PR China; email: 1819011117@stmail.ntu.edu.cn, 1819011088@stmail.ntu.edu.cn. All authors collected the data. Bing-Bing Wu extracted and analyzed data and wrote the original draft. Dong-Zhou Gu and Li-Ping Feng extracted and analyzed the data. Jia-Ning Yu processed the tables and figures. Rong Xu interpreted and reviewed the data. Man-Li Zha and Wang-Qin Shen revised the article. Wang-Qin Shen and Hong-Lin Chen conceived the study design and content concept.

The study was funded by Nantong University College Student Innovation Training Program Project 2021 (Number: 2021163) and the Basic Scientific Research Project of Nantong, Jiangsu Province, PR China (number: JC2020044). Man-Li Zha also provided funds.